This is a major release. It has been in pre-release for quite a while and some of you have already been using it. Please refer to all the V1.11.x change logs to see what changed. Here is the biggest highlights
- We now offer Live Watch for OpenOCD, STLink and JLink gdb-servers. Live Watch allows you watch a few global variables without halting the processor. The interface is similar to the standard VSCode
MATCH
panel. It has only been tested with the SWD interface (not JTAG). This is still considered EXPERIMENTAL. In yourlaunch.json
you can enableliveWatch
(please use IntelliSense and it will give you sensible defaults). You can also follow the development of this in Issue #810. We have had some stability issues with JLink though with some devices. - Our extension has been split up into multiple extensions so it can work with other debuggers like Microsoft, Arm, IAR, etc. They also work non-Arm devices. Please submit issues to the respective repos. We attempt to install these extensions automatically but had some reports of things not installing due to non-standard installation methods. We are sorry for any inconvenience
Note: Please install this extension using VSCode with a working internet connection. Do not try to install it manually. If you install it manually, then also make sure all the other dependent extensions are installed.
- Issue #861: Potential fix
- Issue #867: STLink make it so that user has to enable
-shared
if needed. Before it was automatically added to the command-line and there was no (good) way to remove it - Issue #882: Potential fix. In some cases the extension would crash while single stepping
- Issue #896: Bug in handling of OS specific overrides (like
cortex-debug.gdbPath.linux
)
- Minor bug fix to guard against invalid expressions (watch and live watch)
- Some documentation updates for better IntelliSense
- Bugfix: #853 Multiple SWO (OpenOCD tpiu_swo_trace) Connections for External OpenOCD Server The issue is with very long connection times inside WSL2 where we would timeout and try again. This caused multiple SWO connections which we did not even know was possible.
- Removed all (user visible) remnants of the old disassembly. The new disassmbly view is provided by VSCode and we supply the contents.
- New handling of static variables. This is also a big change and it will be visible in the Variables window near the
Statics
section- You will see the file the static variables are associated with
- When you add a static variable to the Live Watch window, it will show the simple name but on a hover you can see which file it comes from
- We needed this to disambiguate in the
Live Watch
window the origin of the static variable as they may not be unique - If you add a static variable from the
Variables
window to the builtinWatch
window you may see the entire path name. We cannot help it but you can edit it - Known issue: no fix identified. The
Add to Live Watch
was added to all theVariable
window elements. We have no way to tell VSCode, which categories (Scopes) should be allowed. So, on a right-click, you will see this menu option but it will not work and you will be notified if you use it.
- Bugfix: #832 Issue decoding templatized class.... Because of all the
::
and<>
, variable names had to be quoted for GDB. Thanks to the issue submitter @jwbmwv for giving me a very simple test case
- Many features to Live Watch added
- You can now edit the expression
- You can move up/down the expression (they rotate if at edges)
- You can enable the global hex mode or use the hex mode on one expression at a time just like you can do it in the
WATCH
window- To enable hex mode on a single expression, simply add a
,x
at the end of the expression. You can also do octal by using,o
. However, this only works for the scalar variables and if it is a pinter or an array, it is not inherited by its children
- To enable hex mode on a single expression, simply add a
- There is now a button on the
VARIABLES
,WATCH
andCORTEX LIVE WATCH
panels title bar to enable/disable Hex display mode globally. You can still override on individual expressions. - Oops. The context menu in the Variables window for enable/disable Hex had disappeared.
- Increased the number of threads (256) and frames (4K) you can have. Such limits should not even exist but it is an artifact of how VSCode works
- The Registers Panel is now removed. We had deprecated this window a year ago as it would not give you accurate register values based on the Program/Thread/Frame. The accurate version of the Registers have been available in the VARIABLES panel for almost a year now.
- Fixed some issue when the Live Watch window was empty or variables added while there wasn't a debug session sometimes would not show
- Bugfix: Potential fix for Issue#840: Variables mixed with a bunch of configuration constants. Not really a bug but in some cases, the ELF file contained a bunch of symbols that were marked as real object which in fact are not. Trying to filter them out.
- Sorry we had some version number issues. This release was accidentally released as a production 1.7.1 and to undo that, we had to jump two minor versions numbers to keep up with VSCode versioning scheme. Apologies for the confusion.
- This is a preview (Pre-release). The biggest feature is the addition of Live Watch. In your launch.json add an object property called
liveWatch
and Intellisense should fill out a template to enable Live Watch."liveWatch": { "enabled": true, "samplesPerSecond": 4 },
- MAJOR MAJOR change: The Peripherals (SVD) panel is now maintained in a separate extension. We are disabling SVD support in this extension. Please refer to https://github.com/mcu-debug/peripheral-viewer. The new panel will show up as 'XPERIPHERALS'. You don't have to make any changes in your configurations. Things should just migrate over. Fingers crossed
- There are tons of other changes planned as pre-releases and they will have versions 1.7.x (this is a Microsoft convention). There has been some major restructuring of our code and as such, this version may not be as stable as the production releases. Please help us get to production
- To enable Pre-releases, you have use the Extension Manager/Pane within VSCode
- Bugfix: #831 Unable to read variables during 2nd breakpoint stop
- Issue#793 Container install broke from v1.6.7 to v1.6.9. It appears that VSCode does not install extensions with dependencies fully. Ie, if ext.
a
depends onb
which depends onc
,c
is not installed so extensiona
fails to load properly. While we confirm, report this issue to VSCode folks, a (ugly) workaround has been implemented.
- The RTOS view has been moved to another extension so it can work with other debuggers. https://github.com/mcu-debug/rtos-views It will be automatically installed for you with this update. Note that the RTOS view related settings and commands now belong to the new extension (search for
RTOS
in Command Palette) - This release will also install the new memory view extension from https://github.com/mcu-debug/memview. The old memory view is still available (search for
legacy
in Command Palette) - Fix for Issue#777. This supports a new (undocumented) type of watch-point. Depending on version of gdb a watch-point could be called
rwpt
orawpt
, so we not accept both. - Fix for Issue#786 SWO/RTT grapher cannot display
- Minor useability fixes
- PR #761 pre, post and override Reset commands. There are three new launch.json properties to customize the behavior of the
Reset
button. If they don't exist, they will correspond to the previous defaults which is to use theRestart
counterparts - [Issue #772](OpenOCD server: Spelling mismatch for RTOS 'ChibiOS' vs. 'chibios') OpenOCD documentation which is what we rely on, had typos for official name of RTOSes. We had to through their C source code to determine the real ones. Please use
auto
for the RTOS value as it is pretty effective. - JLink now properly supports a
detach
operation so that programs can continue running after the debug session. Please upgrade your JLink SW to V7.82b or later.
- The old memory viewer is temporarily restored as a new command
Cortex-Debug: View Memory (Legacy)
. We will not be maintaining this legacy viewer but there may be capabilities there that are not yet in the new memory viewer. Like selection and copy to clipboard
- #769: Nested register cluster in SVD file is not displayed. The SVD spec changed in v1.3 nested clusters are now allowed. This is now supported by Cortex-Debug
void *
variables now returnmemoryReference
values.- ST-LINK we now reserve four TCP ports when launching their gdb-server. We only use the first one but it done in case a second ST-LINK server is launched.
Cortex-Debug: View Memory
command now uses a separate memory viewer extension if it is installed. https://marketplace.visualstudio.com/items?itemName=mcu-debug.memory-view. In the future, we may auto install the extension when this extension is installed. We would like to hear if you would like that. The download is about 1MB
- Issue #741: Could not determine gdb version from Fedora install
- Issue #730 fixed where bulk breakpoint changes would cause issues and some breakpoint actions would fail. Like enable/disable/delete all breakpoints.
- Mitigation for Issue #725 with extraneous/dummy symbols masking real ones
- Bit more lenient parsing of GDB version number for custom builds. We still need to know the version number
- Better error message when using GDB version less than 9. Starting with 1.5.2, we no longer support GDB version <= 8. For most of last year, we have been warning users of this happening in July 2022. This now August 2022 :-)
- Please see all the notes from the pre-releases (1.5.x). Following are the highlights
- RTOS Views: Support for uC/OS-II & embOS, thanks to @PhilippHaefele & @mayjs for adding them. Just a reminder, anyone can contribute their favorite RTOS if you have a bit of knowledge of TypeScript/JavaScript and more importantly, knowledge of RTOS internals
- You can also use Microsoft Embedded Tools and their RTOS views as we are now compatible with each other
- Microsoft Embedded Tools also added compatibility for Cortex-Debug for their version of Peripheral Views/Registers. They also helped with integrating Cortex-Debug with MS built-in Hex-Editor. To use MS Peripheral View, use
svdPath
in launch.json.svdFile
uses the SVD feature from Cortex-Debug. You can use both at the same time if you wish but you should typically use only one. - Fixed a long standing issue with OpenOCD RTT where there was no good way to know when to start the RTT. We can now poll until RTT is detected and enabled. See
rttConfig.rtt_start_retry
in your launch.json to control who the polling works. - Support for loading alternate symbol files instead of the
"executable"
using a new launch.json property"symbolFiles"
. See the Wiki documentation here. This is in addition to the already existing"loadFiles"
which is used to customizing the programming of the device - You can now use
breakAfterReset
andrunToEntryPoint
for anattach
type launch configuration as well but they will only be used on a reset/restart. - For chained configuration, the parent can override properties of children and/or allow children to inherit from itself using
overrides
andinherits
. - Of course, many issues fixed: thanks to the community with reporting them helping fixing them
- Enables
Peripheral View
provided byEmbedded Tools
extension from Microsoft. You can request this by specifyingsvdPath
property inlaunch.json
.svdFile
still remains and that will display a window named 'Cortex Peripherals'. For now, you can use both but will require twice as many memory reads. Make slow down fast single stepping.
- Version 1.5.3 has accidentally enabled some debug info that is should not have enabled during production. Corrected
- Serial port changes: We have no changed how serial ports are supported. We now rely on the Serial Monitor extension from Microsoft to provide a copy of the
serialport
NPM module or you can compile it for cortex-debug yourself. The latter is always preferred. We do this becauseserialport
is a binary module and requires it to be compiled using the exact version of Electron and Node that VSCode is built upon. This has become very difficult to support for every OS/Processor combination and it was also increasing the size of the corte-debug package -- exceeding the limits set by Microsoft - Cortex-Debug RTOS views will now work with MS
cppdbg
debugger and vice versa. You can use both or either or none. Until the next release, you will need pre-release versions of the Microsoft Embedded tools and this extension. - Thanks to a PR by MS embedded folks, you can now have memory views for pointers from the Variable and Watch windows.
- RTOS View for uC/OS-II. Added by @PhilippHaefele & @mayjs via PR #642
- RTOS View for embOS. Added by @PhilippHaefele via PR #705
- Better error handling during RTOS initial queries
- Memory references are returned from DAP evaluate requests. PR #694
- RTT with OpenOCD will attempt to retry the
rtt start
command if the first attempt fails. This is done as a background task. The retry interval is controlled byrtt_start_retry
option inrttConfig
(default is 1000 ms) - Issue #698: Windows only, fixed path-name issues with
loadFiles
and a couple of other places - PR #700: RTT with JLink can now be used with any valid channel (0-15) but you can only have one channel per session which is limited by JLink server. Before, only channel#0 was allowed
- Issue #689 fixed
- Reset button now resets all cores (using lifecycle management rules) and works the same as Restart
- Support for loading alternate symbol files instead of the
"executable"
using a new launch.json property"symbolFiles"
. See the Wiki documentation here. This is in addition to the already existing"loadFiles"
which is used to customizing the programming of the device - You can now use
breakAfterReset
andrunToEntryPoint
for anattach
type launch configuration as well but they will only be used on a reset/restart. - For chained configuration, the parent can override properties of children and/or allow children to inherit from itself using
overrides
andinherits
.
- #643: Timeout for runToEntryPoint has been removed because it is never right for some use cases and don't need another setting. User can always use the pause button.
- #640: Fixed "Inaccurate showDevDebugOutput hint message"
- #636: Fixed "Chained configurations: 'folder' option is no longer working". Was an issue only on Windows
- #635: Fixed "Debugging hangs until relaunch". Let us know if it is still happening for your use case
- #643: Maybe fixed: "ECONNRESET by J-Link GDB when removing "runToMain": true". Revamped the way VSCode handles the startup sequence. It is VSCode that is confused because of the sequence of events.
- #645: Fixed "SWO support is not being fully enabled in pyOCD"
- #647: Fixed "Serial SWO source with OpenOCD". It was not using the right swoConfig property to open the serial-port. The same problem existed with
external
type servers. - #653: Maybe Fixed "SVD Display Stops". At some point, VSCode TreeView changed and fails on NULL tooltips. Was not a problem before General
- There was a mis-understanding in how we interpreted the VSCode API (we took it literally). This caused some issues with synchronization with Cortex-Debug and VSCode. This caused hard to reproduce issues and perhaps it is addressed now. There is an issue here with JLink as well with creating and deleting threads while program is paused -- this should not happen, but it confused gdb thoroughly. Happens at startup and maybe it masked now.
- Improves simultaneous 3+ core debug. There was an issue where if multiple debug sessions were going at the (near) same time, when we thought a TCP port was free, it truly wasn't because the gdb-server involved was too slow to claim the port. This is more evident in chained configurations, where things happen rapidly but it was more of an issue with JLink because of how that server works.
- Bugfix: #618, #619: Work around a gdb bug in info-variables. gdb either takes forever or crashes or runs out of memory
- Bugfix: Implemented a workaround for VSCode issue with pause not working if neither
breakOnReset
orrunToEntryPoint
is set. - Bugfix: Workaround a gdb issue where it loses track of current thread when the current thread is deleted. This caused gdb
continue
to not work as expected. It may actually be a problem in the gdb-server as it reports a thread being deleted after a breakpoint/pause -- ie, program status changes while paused which is not expected.
- Bugfix: Mistake in the release. No significant change
- FreeRTOS View performance improvements
- FreeRTOS View now provides FW hints for missing information
- Experimental: A memory view provided by VSCode can be launched by clicking on variables in the Variables Window. When available it is indicated by a small icon on the right edge.
- We don't yet allow this for all variables. Only those that are obviously pointers. So no arrays, structures, expressions, etc.
- VSCode is not very nice at the moment. It requests 128KB of memory (twice for one pointer) which for a lot of people could take a lot of time and it affects every time you step/pause/step. We will work with Microsoft to see if this can be customizable. If not we will remove this feature.
- There are some issues when the debug session ends -- like it wants to save the file to disk. But only in certain circumstances
- Bugfix: The RTOS info is more accurate and it was failing earlier when trying to calculate Peak stack usage. The headers are now two rows and all columns are always displayed -- letting you know what else you can see provided you enable those features in FreeRTOS. The table is now more compact and colorful.
- Hover is now much more powerful. You can expand arrays, objects, etc.
- RTOS View is now available. It is experimental and there is only support for FreeRTOS. With your help, we can add support for more RTOSes. You can enable this feature using the extension setting
"cortex-debug.showRTOS": true
via the interactive Setting dialog or adding it to User/Workspace/Folder settings.json. You can also use Command PaletteCortex-Debug: Toggle RTOS Panel
command. See #605 and here is a screnshot
Some notes- It can take a bit of time to update the RTOS info as it requires many tiny queries from GDB. This only occurs if the RTOS panel is visible. You can hide (expose/focus some other panel like
DEBUG CONSOLE
orTERMINAL
). This will prevent any GDB traffic/updates and won't interfere with fast single stepping or waste your CPU resources. When when exposed it will immediately update itself and continue to do so as long as it is visible - If you are not using an RTOS, there is near zero overhead even if the RTOS panel is visible. The lack of an RTOS is quickly detected using one tiny GDB query and all activity ceases thereafter
- The number of columns can vary depending on what features you have enabled disabled in FreeRTOS. For instance, the
Runtime
column will not appear ifconfigGENERATE_RUN_TIME_STATS
was not enabled in your FreeRTOS config file -- which may require a timer faster than the RTOS tick. - We tried to follow VSCode themes and hopefully it works well under all themes and color schemes.
- Note: Microsoft has created a similar feature/extension but it is hardcoded to work with their
cppdbg
debugger. It can easily work with ours and when it does, we have to evaluate which one is better. We have offered our help but got no response yet. - TODO:
- Tie a memory view to examine stack data. Helpful? Not sure.
- Add support to examine queues, events, semaphores, etc.
- Maybe some performance improvements
- Prettier -- I am no HTML/CSS expert and appreciate if someone can help
- It can take a bit of time to update the RTOS info as it requires many tiny queries from GDB. This only occurs if the RTOS panel is visible. You can hide (expose/focus some other panel like
- See logs for pre-releases V1.3.x
- STLink: We no longer use
monitor halt
during startup because halt is automatic with a --attach or --halt command-line options. Also,monitor halt
is not supported in some versions of STLink server. We also now use --attach when attaching as documented in the STLink manual. Most people will not see much difference but it helps when people are using older versions of the server. More importantly we do what the STLink manual suggests. - Initial support for
overrides
andinherits
in chained configurations. - Issues addressed: #599 #598, #539, #295, #601, #602, #593
- ST-Link users may see a change a couple of changes
- In version 1.2.X, a change was made to automatically add certain directories to LD_LIBRARY_PATH (linux) and DYLD_LIBRARY_PATH (Mac). While that worked for most people, it can have un-intended consequences. On all platforms we now use the location of the gdb-server as the current directory to make the executable find the dynamic libraries properly
- In previous versions, when auto-discovering ST link install directories for the gdb-server and the programmer, we picked any installation. ST seems to ship multiple versions, so we now pick the latest version as indicated by the version number in the directory.
- The safest thing to do is to set the paths yourself using one of the extension settings or settings in launch.json. Normally, not an issue but can cause problems when you have multiple installs or installs in non-standard locations.
- Bugfix: Issue #596 If neither
breakAfterReset
orrunToEntryPoint
were used, the pause button failed to work. The program would run but the debug buttons didn't do anything. We don't know which version of VSCode broke this but we created a workaround that is being tested
- Bigfix: Regression since introducing chained configurations a month ago. We were not waiting for the gdb-server to open a TCP port and launching gdb far ahead or time. Most times, this worked okay but this is wrong. Reverting back to old/proper behavior.
- Major change: We now require both
objdump
andnm
for extract the source-file, type, address and size of symbols from the executable elf file.objdump
gives us better (not perfect) symbol information and its types but bad file information.nm
gives good file information but is wrong in classifying symbols types (function vs data). Gdb does not give us size information, hard to correlate and very version dependent. We need everything addresses, sizes, names and origin (files). Without this, we have trouble giving a proper experience wrt Statics, Globals and other things like disassembly.
There is no special setting for this. We modify the path toobjdump
and expect to findnm
there. If not, we will continue with the debug session but experience will not be as good (you will see a warning in the Debug Console) objdump
is not exactly fast and the executable sizes are growing.nm
is even slower. We now changed the startup process to runobjdump
,nm
andgdb
at the same time to improve the startup times. They are now better than before when things were done sequentially. We also launch the gdb-server (like OpenOCD) in parallel as well- The above two are major changes that affect the startup procedure.
- Minor improvement in disassembly.
- Issue #592: We were not properly detecting used TCP ports causing failures in gdb-server starts. On some systems
- Issue #585: RTT Terminals were not being re-cycled. Instead, new ones were being created requiring manual cleanup.
- Improved startup code for
launch
andattach
. Quite a bit of old unneeded code removed following VSCode's current APIs. We were doing a few things inefficiently and thew updated VSCode APIs helped. We have a few too many options for startup likerunToEntryPoint
,breakAfterReset
, etc. along with user defined overrides/pre/post commands. These are consolidated. This will also reduce the number of updated that happen to the various windows. We tested best we could but this is yet another major change. - Also the reset/restart processing uses virtually the same code as startup.
- More improvements in disassembly, faster and better. Tried to make it a bit more generic to handle non-ARM architectures, but 32-bit ARM remains a priority and that is the only thing available for testing.
- Issue #585, hopefully addressed.
- New Feature: Support for Logpoints. Log points allow you to print debug information to the Debug Console without stopping the program (but a breakpoint has to be used internally by gdb). Setting Logpoints is similar to setting/editing breakpoints in the editor. The value of what you enter in the dialog box is similar to arguments to printf (but don't actually use the word printf and NO commas and arguments should be separated by a space)
The above turns into the equivalent of
"Value of counter is %d\n" counter
Any variables that are referenced must be in scope of the log-point and not optimized out by the compiler. This feature uses theprintf("Value of counter is %d\n", counter);
dprintf
feature of gdb. See https://doc.ecoscentric.com/gnutools/doc/gdb/Dynamic-Printf.html and is configurable - Feature: Some refinements in disassembly. If the program ever stops where there is NO source code available, a disassembly is displayed automatically. But we now use the new disassembler which is full disassembly of the program whereas before it was only function disassembly. This is another step towards obsoleting the old disassembler.
- Feature: Allow
runToEntryPoint
forattach
type sessions. Only applies when you use the Restart/Reset buttons - Under the hood: We now use
nm
along withobjdump
to get information of global/static variables/functions. Objdump is unreliable for file names and nm is unreliable for symbol types, gdb is too slow and incomplete. This information is used to display globals/statics in the variable window and for disassembly. There is no setting fornm
but we take the path forobjdump
and replace it determinenm
. It will work withoutnm
being installed but your results may not be as good/accurate. - Bugfix:
Detach
while a program was running was broken under certain conditions. Note thatDetach
is not the same asDisconnect/Stop
.
- Hotfix: Issue #585 -- RTT terminal windows not getting reclaimed/cleaned-up
- Hotfix: Issue #579 -- debugger won't start (hangs) if a program stopped in a function with no source file. Wrong version of the disassembler was being invoked.
This is a rather large release. Contains many enhancements, new features and bug fixes
- Release policy: Our release policy and versioning system has changed. Microsoft recently released the ability to push pre-releases to the marketplace. With that came a new versioning scheme we have to follow. Our version numbers follow the SemVer specification major.minor.patch format. Releases with even minor numbers are final releases, and odd minor numbers are pre-releases. Between the last final release and this one, we have done 11 pre-releases and thanks to those who subscribed to those and helped refine the extension. See https://code.visualstudio.com/updates/v1_63#_pre-release-extensions
Note: We will no longer be publishing releases (or pre-releases) via github as they are automatically pushed to your VSCode and for those that need a VSIX file, you can download it from the marketplace. This is until we can automate the github releases because it is error prone rework to do two kinds of releases - Multi-core, multi-session support: We now have support for simultaneous multi-core debugging using a single instance of VSCode. While you could always use VSCode's
Launch Groups
feature, we have introduced what we term chained launches that will give you more control over sequencing of the launches. More details here and here is a summary- Chained configurations with configurable delays, event based launching and life-cycle management
- You can have a shared server with the TCP/IP ports shared automatically. Both OpenOCD (shared) and JLink (separate) type gdb-servers are supported
- The launch configurations can even be in different devices/boards
- Lifecycle management for start/stop/reset/restart/terminate/disconnect. Sorry, no synchronized stop/step as that is better done via CTI (Cross Trigger Interface)
- SWO and RTT are now session aware
- SVD files: You can have multiple SVD files (one for each session) but hopefully, you only need one as they are slow. But given that we also support multi-session, we do not restrict the number of SVD files.
- One can see the relationship of the various items in the chained configurations in the
Call Stack
window. Children are nested (indented) a bit.
- Full Disassembly with inline source: Thanks to a Viewer from Microsoft and our backend+gdb, we now have full disassembly of your entire program (virtual and on demand so it is performant). See details here. Thanks to @hongshui3000 for taking this for a ride and providing valuable feedback. Note: The old style disassembly of functions without source will be DEPRECATED
Known issue: Sometimes you may not see source code when the current instruction is near a bunch of other functions that do not have source. This happens for example if you are stopped in main, and your main is small, and it is surrounded by startup code that does not have source. Not sure why gdb is doing this - Registers:
Registers
are now available in theVariables
panel. You can now change the values of registers. But more importantly, unlike theRegisters
panel, this will track your current thread/frame in theCall Stack
window. This means that registers are now shown in the context of the current frame. What we had before in theRegisters
panel was information we had at the time a halt/breakpoint was hit and potentially incorrect when you refreshed -- this was because there was no API in VSCode for extensions to track the Call Stack window.The old Registers panel will be DEPRECATED
- Website changes: Our github repo always had a Wiki but it was pretty weak. Many thanks to @PhilippHaefele for a lot of edits he did over the last couple of months and also helping closing many issues that were already addressed.
- SWO configuration: SWO was a hit and miss as multi-core devices appeared and device manufacturers were not using the default base addresses for the ARM debug hardware like TPIU, DWT, ITM, etc. We factored this out in a user settable gdb script as well a small TCL file for OpenOCD which needs additional configuration. See https://github.com/Marus/cortex-debug/wiki/SWO-Output#swo-configuration
- Many small enhancements:
- Platform specific settings for
objdumpPath
- On a restart, now the debugger will stop at your
runToEntryPoint
setting orbreakAfterReset
or just continue execution if neither is enabled. - File extensions
.S
and.s
are recognized as valid assembly source code for breakpoint setting
- Platform specific settings for
- STLink restart and attach were broken
- With the new release of STM32 software, SWO was broken. Should be working now
- Issue #561 more graceful handling when gdb is wedged
- Issue #538: Fixed bug SVD internal debug verification. Not supposed to be for production but got released and caused false errors. This in turn resulted in SVD load failure.
- Issue #522: Qemu launch failed because it does not have a matching regular expression that indicated a start. It never does and code to handle that did not work. Fixed.
- Issue #539: Using GDB to get some symbol information for locals and globals. Hopefully, gives better performance for large executables. Most information still comes from objdump though.
- The
Adapter Output
window in theOUTPUT
tab is no more. We have had a deprecation notice for months now and have been using thegdb-server
tab in theTERMINALS
tab.
- Bugfix: Unable to delete instruction breakpoint
- Print memory region found/used during disassembly (will be removed once disassembly matures)
- Allow file extensions
.S
,.s
as valid assembly language extensions for setting breakpoints
- Feature: you can now customize the
objdumpPath
like you can thegdbPath
you your settings or in launch.json - Some improvements in disassembly
- Maintenance: We were using very old API definitions from VSCode. We switched to the newest APIs for all types of breakpoint (function, file/line, data, instruction). Fingers crossed!
- Bugfix: Double clicking in the disassembly window will now take you to the corresponding source file/line if any
- There have been many other bug fixes for disassembly and a few enhancements. Improved performance, be able to see
- more than one line of source (max of 3) -- note that gdb does not provide more than one line in the API
- If you do use disassembly, there is some extra debugging output in the Debug Console to be able to track things
showDevDebugOutput
is going through a change so we can better debug this extension and users can provide better bug reports. It is now a set of choices.
- Feature: New Disassembly mechanism. While debugging you can right click in the editor or the
CALL STACK
window on any thread/frame and you will see the Disassembly window show up. It is helpful to have the original source code and the Disassembly window side-by-side. THe Call-stack, Register in the Variable Window, C/C++ source code and Disassembly all track. Note that the Disassembly window is managed by Microsoft/VSCode.- You can set breakpoints in the Disassembly window
- If the focus is in the Disassembly window, then you can single step at an instruction level and follow long in both source and assembly. Some scrolling issues may be visible in the window that does not have focus.
- See https://github.com/Marus/cortex-debug/wiki/Disassembly-Debugging
- https://github.com/Marus/cortex-debug/wiki/images/disassemble.gif
- Looking for feedback
- Feature: New property
loadFiles
forlaunch.json
which can provide a set of files that need to be programmed instead of the defaultexecutable
. This list can even be empty in which case, nothing will be programmed. The files can be elf, bin or hex files. This touched every gdb-server and we tested with JLink, openocd, pyocd, stlink and stutil -- those are the ones available to us. - Bug fix: More of an enhancement, avoid using TCP port numbers allocated by the first server for other servers launched during chained launching.
- Chaned launch configurations are now possible where the primary launch type is an
attach
- Bugfix: Issue #561 more graceful handling when gdb is wedged
- Bugfix: data breakpoints could not be deleted. Fixed by @PhilippHaefele
- Feature: There now a command (default key-binding Ctrl+Shift+X) to toggle Hex display in the Variables window. Does not affect the Registers window (which may be going away) as it has its own command and button. You can find this command in the "Command Palette" as "Cortex-Debug: Toggle hex display in Variables window"
- Fixed issue with with SWO (or RTT) not working the first time. It was a race condition where we were trying to connect too fast. Now we re-try. The max timeout is 5 mins.
- Fixed SWO and RTT are now session aware. As in, you can have multiple RTT/SWO windows from different debug sessions.
- BugFix: Reset was broken in previous release
- Reset and Restart should be working now for chained configurations
- Workaround: ST released there latest update to their IDE and along with it came a new OpenOCD and perhaps FW startup sequence. With that, SWO is broken for all releases. We no longer have permission to set the registers required to make SWO work with OpenOCD. We think we found the issue and a workaround (ugly) that is temporarily in this pre-release.
- Known problems:
- If there is NO
breakOnReset
and NOrunToEntryPoint
-- meaning the program starts and continues executing with NO breakpoints, you may not be able to pause the program. However, if you reset/restart or do a fresh start, then things seem to be fine. This occurs only the first time you run a session after launching VSCode. Does not happen with non-ST devices and even this is new. The cause is unknown. - SWO may not work on the first time after VSCode launch. This may be un-related to the problem above. Fixed
- If there is NO
- Registers xPSR and CONTROL now have fields shown in the tooltip. This is in the VARIABLES Window
- Bugfix: There was a bug dealing with large integers (bigger than 32 bits) in tooltip generation. May have caused Register values to be not displayed at all.
- Multi-core debug: Synchronized launching/terminating working, Registers & Peripherals working. Restart/Reset needs a bit more work. The top level configuration has to be a
Launch
session but others can beLaunch
orAttach
. Same will be available withAttach
at the root of the chain --Attach
requires all the configuration to be duplicated, so we wanted to wait for a bit more maturity. - Included with multi-core is multi-session where you can have multiple gdb-servers running at the same time. This is required for JLink where you may have one JLinkGDBServerCLExe connected to one core where as with OpenOCD and others, you will see the same GDB server shared for all the cores.
- With the above changes, you may see multiple
Debug Consoles
,gdb-server
Terminals, multiple programs in theCall Stack
window, multipleRegisters
and multiplePeripherals
. - DEPRECATION NOTICE: We are moving the registers window to the VARIABLES Panel (new category with Locals, Statics, Globals). As much as we liked having a separate panel for Registers, it was not possible (due to a VSCode limitation) for it to track the current Thread/Frame in the CALL STACK window. As such, you may have seen in accurate display. In the VARIABLES Panel, all values are in sync. with the CALL STACK window and it is also more efficient.
- SWO: The method of configuring SWO has changed. There is a new GDB-script that we encourage users to customize. Frequently, vendors are using non-standard addresses for ARM debug components (CoreSight) and hence it is not possible for to detect or keep track. You can now customize the script yourself. See https://github.com/Marus/cortex-debug/blob/master/support/gdb-swo.init. Currently, only OpenOCD uses this script while the others are still using the old scripts that were not easily modifiable. Feedback, improvements welcome.
- OpenOCD requires additional configuration that is now done in a TCL script (config file in OpenOCD lingo). This is also user customizable as vendors are using non recommended ways of defining SWO/TPIU configuration. This is user configurable. See https://github.com/Marus/cortex-debug/blob/master/support/openocd-helpers.tcl. This file also has what is required to configure for an RTOS as the previous hard-coded method does not work well for multi-core systems and for those who do not follow OpenOCD recommended naming conventions.
- For STLink Gdb-server and Cube-Programmer, set
LD_LIBRARY_PATH
on Linux and equivalentDYLD_FALLBACK_LIBRARY_PATH
on MacOS - STLink: We now use the
--halt
command-line option when launching the server unlessserverArgs
is used. IfserverArgs
is used, you can enable the--halt
if that is the behavior you desire. Things may not work in the future without the--halt
option. - STLink: We no longer use
monitor reset halt
as that is not a supported reset option. We use the--halt
command-line option instead. - Experimental:
runToEntryPoint
is now supported forReset
andRestart
. This was not possible/recommended before because of how VSCode handled breakpoints. SincerunToEntryPoint
works by setting a temporary breakpoint in the specified function, this may fail if your CPU has run out of breakpoints. - Experimental (Unstable): Chained configurations for multi-core/multi-processor/multi-session debugging
- Bug fix: When using arm tools version 9+, some global variables were not being displayed
- Version numbering change. Now that VSCode extensions are allowed to publish pre-releases via the marketplace, we will be moving to a new version numbering schema. A version number is (following SemVer standard) major.minor.patch. The general SemVer standard also allows a suffix for alpha, beta, etc., but this is not supported by the marketplace. Instead, they recommend that ODD minor versions be pre-releases and EVEN ones for release versions. This versioning scheme may change in the future. See: https://code.visualstudio.com/api/working-with-extensions/publishing-extension#prerelease-extensions This will be the first pre-release coming to you direct from the marketplace.
- Issue #538: Fixed bug SVD internal debug verification. Not supposed to be for production but got released and caused false errors. This in turn resulted in SVD load failure.
launch.json
optiondemangle
is removed. We always demangle. Its default value was true for quite some time. There is quite a bit of C++ now coming in and Rust as well. It appears there is no harm in always enabling it.- Issue #539: Using GDB to get some symbol information for locals and globals. Hopefully, gives better performance for large executables. Most information still comes from
objdump
though. - Issue #522: Qemu launch failed because it does not have a matching regular expression that indicated a start. It never does and code to handle that did not work. Fixed.
- Issue #536. Typo in code for os-specific paths
- Store register/peripheral settings in the appropriate folder instead of the first folder
- Kill gdb-server if the user kills/exits gdb without using the proper disconnect/Stop buttons/process/commands
- VSCode was terminating Cortex-Debug before it was done. st-util exit behavior was also not clean as it did not exit on a disconnect.
- Preliminary support for data watchpoints
- SVD now can allow no merge of consecutive addresses with a -1 specified for
svdAddrGapThreshold
. This will make peripheral updates very slow but certain devices may need this option. - You can now save the output of a gdb-server into a text file by specifying
cortex-debug.dbgServerLogfile
in User/Workspace/Folder settings. This will save output from the servers from all sessions (not just the recent ones). This is primarily for debugging and for users when submitting issues. - Path names for gdb-servers can be OS specific. For instance
cortex-debug.openocdPath
can be suffixed with one of.linux
,.osx
or.windows
. For instancecortex-debug.openocdPath.windows
is used only on Windows and if that is missing, it will default looking for cortex-debug.openocdPath`. - SWO output can now be logged (saved) to a file just like RTT output
- Issues #524 and #525
- Improved handling of J-Link RTOS option (file extension), added NuttX
- Fixed a regression for STLink gdbserver. It was in fact accidentally working in prior releases. The real bug is now fixed. Issue #494
- We may have finally found a way to exit OpenOCD without having to kill it and OpenOCD not hanging around after the session ends. This is of course dependent on OpenOCD behaving as documented. Thanks to #482 and @bohdan-tymkiv for a solution
- Timestamps for RTT and SWO have been standardized to be of the form
[ISO-Date-Time, +NNNNNNms]
where the first part is the date/time of day and the NNNNNNms is the number of milliseconds elapsed since the debug session began. timestamp
is now an option for SWO console decoders. Default isfalse
. A timestamp is output only when a newline is received or a timeout of 5 seconds
- Bugfix: Issue #493 In the previous release, we were trying to end OpenOCD using a SIGINT first and then SIGTERM. The way VSCode works, this did not work in production releases. Reverting back to the previous method of just using SIGTERM. Unfortunately. Still looking for a better method to end OpenOCD.
- Support for resume/suspend after Launch/Attach. With new UI features added to VSCode, the Stop button (after
Launch
) can now also be used for a Disconnect using keyboard shortcuts. The reverse is true when using anAttach
type session. But this requires co-operation from the gdb-server to comply. Certain versions of OpenOCD do comply, JLink always seems to resume (see issue $481). Provided the gdb-server cooperates, the expected behavior now when you end a debug session is:Stop
will leave the program in a halted stateDisconnect
will let the program continue
- You can now have RTT console output lines contain a timestamp. Use the
timestamp
option for the RTT decoder. Default is false (no timestamp) - Issues #482 addressed for JLink. It always cleanly exits on it own. OpenOCD is still an issue where it has to be killed which once in a blue moon does not seem to work.
- Integrated PR #480 -- creates .vscode director if it doesn't exist for saving register/peripheral states.
- PRs #489, #490, #488, #478 submitted by @trond-snekvik merged. They fix issues and enhances your experience with Cortex-Debug in various ways. Thank you @trond-snekvik
New Features
-
Reset button: There is now a button for resetting the device in the Debug Toolbar. This is distinct from the
Restart
button provided by VSCode framework. TheReset
button does not trigger a rebuild but other than that, the behaviour should be the same. The customization commands forRestart
are also used forReset
. Those arepreRestartCommands
,overrideRestartCommands
,postRestartCommands
andpostRestartSessionCommands
.- The only place we could put the
Reset
button was at the beginning. We would have rather put it closer to the end but there isn't an API to do that as far as we know. Note that the toolbar appears in docking and floating modes and it is the docking mode where this was not possible. Hopefully, it is not a hinderance. Restart
causes VSCode to clear theDebug Console
butReset
does not do that as VSCode is not involved.- Some gdb-servers do not respond appropriately to being reset and are not compatible with
gdb
expectations. You will have to find out what set of gdb-server/gdb commands work for you and use appropriate options in thelaunch.json
. For instance, just after 'reset halt' some versions of OpenOCD do not providegdb
the updated registers (SP, PC, LR, etc.). Some devices may have a more complicated, customized reset mechanism, boot-loaders, etc. To force a synchronization betweenOpenOCD
andgdb
, you can do the following inlaunch.json
.
- The only place we could put the
"postRestartSessionCommands": [ // OpenOCD only
"monitor gdb_sync",
"stepi" // Because of the command above, this is a no-op. No code is actually executed
]
-
Run Without Debugging (^F5)
: Experimental. This will now work but VSCode does not clearly define what this button should do. In an embedded cases, that is even murkier because without GDB and a gdb-server, there is no way to start the program. Between VSCode and Cortex-Debug, the end result is as follows- VSCode does not transmit any breakpoints to Cortex-Debug and hence no breakpoints
- VSCode does show a pause button in active mode but pressing on it does nothing because that action is not sent to Cortex-Debug
runToEntryPoint
andbreakAfterReset
options are disregarded- If the program halts because of an exception or any other reason, it is handled normally and now you will enter the normal debugger
-
Auto-continue: New behavior. Operations
Launch
,Reset
, andRestart
will now issue acontinue
to gdb upon successful reset-halt. This is not done in the following casesrunToEntryPoint
has been used for aLaunch
session or it is anAttach
session- If a post-session-start commands (i.e.,
postStartSessionCommands
,postRestartSessionCommands
) are used; you can insert thecontinue
command in there. - Or you have used the
"breakAfterReset" = true
-
svdAddrGapThreshold
option inlaunch.json
. Normally adjacent register addresses with small gaps are combined to reduce the number of (slow) device memory reads. Default is 16 bytes. You can now control the number of bytes the gap can be including zero. Zero means strict reading of bytes but adjacent registers are still combined. -
JLinkGDBServer will no longer display a graphical progress bar at
Launch
. If you need it, you can use the-gui
command-line option inlaunch.json
'sserverArgs
- Registers (CPU and Peripheral) now indicate with a highlighted value, which ones changed since last update
- Line-based Breakpoints now visually indicate which line a breakpoint is actually set when different from what was requested. Normal VSCode behavior is to revert back to the original line when debug session ends.
- Peripheral update will try to update as much as possible instead of bailing entire peripheral update after a single memory read failure. Failed reads are now indicated with
0xffffffff
-
Now you can toggle Hex mode for Registers and Variables independently from the Debug Panel.
-
The title bar of the Registers View contains a button to toggle between Hex and Natural modes
-
In the Variable Window, you can right click on any variable and it allows you to toggle the Hex mode. This setting applies also applies to Watch variables. Too bad we don't have access to the title bar and there is no good way of implementing a per-variable format
-
-
There is now a Refresh button in the title bars of the Registers and Peripheral Windows. The registers window, when refreshed will use the current stack/frame to retrieve the values.
-
You can hover over a scalar variable name and get a tool-tip that gives you decimal, hex, octal and binary forms of the same.
-
Terminal input
- Support for
Paste
in RTT in RTT terminals (See Issue #463) - Input entered into RTT Terminals can now have their encoding be use using the
iencoding
option for theconsole
andbinary
encoders
- Support for
-
Global variables are now sorted in the Variables Windows. But all variables starting with double underscores
__
are pushed to the bottom.
Minor bug fix. The launch.json
option clearSearch
was not working for rttConfig
. Only affected OpenOCD users.
This is a major release with a lot of changes and many new features. The TERMINAL
area of VSCode is utilized a lot to enable bidirectional communication with the firmware. It is used for RTT, SWO and Semi-hosting.
New Features:
- Support for RTT (SEGGER Real Time Trace) with OpenOCD and JLink. This RTT host side implementation has the following features. See https://github.com/Marus/cortex-debug/wiki/SEGGER-RTT-support for more details
- If you are used to RTT tools from SEGGER, this implementation adds many features, especially if OpenOCD is your gdb-server
- Setup of RTT is automatic by default. For this to work, your executable needs to have symbols so we can locate the address of the global variable
_SEGGER_RTT
- The start of the RTT control block contains a string that OpenOCD/JLinkGDBServer look for. If the address auto-detected, we clear out the string. This will help with cases where you might have stale information from a previous run.
- For OpenOCD, you can customize the
polling_interval
, and the search string. The defaultpolling_interval
is 100ms as of today. 10ms seems more acceptable as a tradeoff between creating bus traffic and not losing/blocking data. If nothing changes in the MCU, then OpenOCD does not do much even if the interval is small. - It is perfectly fine to have Cortex-Debug enable RTT but not use any display features. This way you can use external tools (like JLink tools or custom ones)
- You can plot RTT data just like you could with SWO. The setup in launch.json is identical. See this comment.
- Channel sharing: You can use the same RTT channels in multiple ways. Cortex-Debug reads the channel data from OpenOCD/JLink once and distributes to all subscribers (terminals & graphs & logfiles). For instance, you can plot a channel and also look at its binary data in a terminal. Just use two decoders with the same channel (actually called port) number.
- Note: JLink RTT has a limitation that it can ONLY work with one channel (channel 0). There is another artifact with RTT channels where you may see output from a previous run at the very beginning.
- Note: This implementation does not support Virtual Terminals that you see in the JLink RTTViewer. All output goes to the same terminal.
- SWO console and binary decoded text data now appears in a "TERMINAL" tab instead in the "OUTPUT" tab
- All gdb-server (OpenOCD, JLink, etc.) output is also in the "TERMINAL" tab. In there you can also interact with your semihosting
- The terminals for all the features above have the following (optional) features
- RTT and gdb-server terminals allow user input. Used to communicate with your FW
- Set the prompt to be used (including no prompt)
- Set label used for the terminal. This label is used to the far right where you can switch between terminals
- In VSCode yuo can now drag a terminal window to the editor area or to on of the panels on the left
- Your FW can emit ANSI escape sequences to set colors, font attributes, etc.
- The terminal supports various
inputmode
s. The terminology was borrowed from age old Unixstty
cooked
- Default: Line mode. This is the normal mode. Bash style line editing is supported. In this mode, the FW will not see your input until you press Enter/Return key.raw
- Zero input processing is done. Keys are sent to the FW as you type. The FW has to do any input processing it wishes to do. Not even backspace will workrawecho
- Minimal processing is done. The terminal will echo characters you type and as such handle the Enter/Return keys as well. However, input is still sent to the FW as you type with zero processingdisabled
- No user input is allowed. This is useful when the FW is not expecting any input and for unidirectional Terminals.
demangle
is on by default. You can turn it off inlaunch.json
- Support in debugger for
Jump to Cursor
, thanks to PR#417 - A change in this pre-release, you will see some debug information in the gdb-server console. You will also see messages output by the extension that are not part of the actual output in bright magenta. This will happen in all terminals (RTT, SWO and console)
- WARNING: The
Adapter Output
window in theOUTPUT
tab will go away. Replaced by the 'gdb-server' in theTERMINAL
tab mentioned above.
A big thanks to Henrik Maier @hwmaier for helping me with the RTT feedback. Without that, it would have taken a lot longer and perhaps not as nice.
New Features:
"external"
server type now supports SWO. It works in the following way depending on"source"
"source": "probe"
-- We do not recommend this method as there are issues with File I/O buffering causing delays and stalls, especially on Windows. Use a"socket"
instead. It will use a auto-created temporary file name.- On Windows, it will use normal file-io
- On Linux/Mac, it will use an OS supported FIFO which is more efficient than a normal file
"source": "socket"
(best option if available)- You MUST specify the
"swoPort": "[host:]port"
option in the"swoConfig"
- You MUST specify the
"source": "file"
, then the file specified by"swoPath"
will be used. Same caveats as when"source"
is"probe"
but you have control over the file-name
"openocd"
server type will now use a TCP port for SWO instead of file/fifo for a more reliable connection across all platforms. Port selection is automatically. You can still use a serial port in which case,"source": "serial"
.- Support for
pyoocd
SWO over a TCP port. You can specify the SWO source asprobe
. This is rather new forpyocd
, so it hasn't been extensively tested.
Bug fixes and minor changes:
- Use the
pyocd
executable withgdbserver
as first argument instead of thepyocd-gdbserver
executable. This is a potentially breaking change but long overdue as the old use model has been deprecated bypyocd
. - Few other minor changes
New Features:
- Added a new
runToEntryPoint
launch.json
option that takes a configurable name for the entry point instead of assuming main. This change deprecatesrunToMain
(whilerunToMain
will still function currently it will likely be removed in the future; it is recommended to replace"runToMain": true
with"runToEntryPoint": "main"
in yourlaunch.json
configurations). This addresses issue #389 - thanks to manuargue for yet another contribution to the project.
Bug Fixes:
- Fixed issues with P&E Micro GDB support when using SWD connection to the target - thanks adamkulpa for the PR.
- Fixed issues with being unable to set breakpoints in rust, assembly, and cortex-debug disassembly views on Visual Studio Code version 1.53
New Features:
- Enable ChibiOS RTOS support for the J-Link server
- Added additional details to the the register and field level hover tooltips in the peripheral register view.
This feature upgrades our VSCode dependency for the extension - as a result V0.3.10 will only support the Visual Studio Code version 1.52 and above.
Also wanted to call out the gdbPath
user setting that was introduced at V0.3.8 (but not included in the changelog) - this allows users to override the complete path to GDB including the executable name - allowing the use of gdb-multiarch
.
New Features:
- Added support for Linux ARM hosts (both aarch64 and armhf) to the binary modules
- Added the ability to pin certain peripherals in the Peripheral Register view so they remain at the top of the view. Thanks to manuargue for the PR. This provides an alternate solution to issue #370
- Added the ability to set
gdbPath
as an override in your launch.json file.
Bug fix
- New Feature:
- Added initial support for STMicroelectronic's official ST-LINK GDB server. This server is currently only supported on Windows as we are having difficulties with getting it to run inside the visual studio environment on Linux and Windows. Also, this GDB server does not currently support SWO output through the probe. Big thanks to hwmaier for their PR that provided most of this support.
- Added a
gdbPath
user setting (settings.json) - this allows you to override the full GDB path including the executable name - allowing the use ofgdb-multiarch
for example. Note that currently this only exists in user settings.
- Issues Fixed:
- Updated binary modules to support current versions of VS Code, fixing the support for serial based SWO output - some older versions were also removed to download size. Currently this is supported on macOS (x64), Linux (x64) and Windows (x64, x86); support for Linux AArch64 and armhf architectures is coming soon.
- Fixed Issue #382 - GDB commands in configuration (such as postLaunchCommands) that had had quotation marks (") were not handled correctly when being passed to GDB
- Fixed Issue #374 - Breakpoints are not cleared in GDB when exiting a debug session - this can cause issues in GDB servers like J-Link that support flash breakpoints as they would not have a chance to remove the modifications made to the flash memory to support that feature.
- Fixed issue with not being able to properly disable function break points
Minor bug fix release
- New feature
- Support for IAR produced elf files which just did not work with V0.3.6. They didn't quite work with earlier releases either. They are non-standard (kinda) compared to what gcc produced or what objdump documents
- With the latest version of VSCode, there were too many popups coming. They are all harmless, but annoying. There were changed made to suppress these but they caused other problems. We may just have to live with those popups (that are non-modal and disappear after a while) until we figure out how VSCode has changed. May take a while
- Issues Fixed
- Fixed Issue #273. Handle peripherals blocks with no addressBlocks. Also fix issue with multiple address blocks where only the first one was being used
- Fixed Issue #284. Implement/support 'derivedFrom' at register level.
- Fixed Issue #306. Support displaying more than 20 stack frames
- When runToMain was enabled there were 1-2 harmless popups when the program stopped in main. They were very frequent on Windows, less frequent on Linux and very rare if any on a Mac.
Minor bug fix release
- New feature
- A performance improvement has been made at startup. It was taking 2-3 seconds to parse the symbols out of the elf file. It is now cut in half and the results are cached. When cached results are used, it takes a few milliseconds.
- Issues fixed
- Fixed Issue #263. Static functions were not properly detected because of the differences in how gdb and objdump reported pathnames. It made things like step/next to not work. Much thanks to @Lykkeberg and @stalyatech for reporting and testing solutions
- When using multiple processors, TCP ports should be allocated consecutively, remote chance of failures avoided
serialport
binary module updated for latest Node.js/Electron- Watch and hovers caused too many popups. Something changed in VSCode. These are now suppressed. There were also popups when setting a breakpoint while program was running and single stepping too fast and these are now suppressed.
- When
runToMain
was enabled, it caused a popup to appear for a failed stack trace. Again something changed in VSCode where it is requesting stack traces when program is not stopped.
This is a pretty big release. The biggest change is to address C++ (and maybe Rust) de-mangled names. It had a big effect on the overall code base especially with regards to how disassembly was provided. The separator ::
caused quite a few issues and there are strange and unexpected things in the symbol table. It can affect users not even using C++ or de-mangling.
- Issues Fixed
- Issue #232: SVD Enumerated types
derivedFrom
attribute now supported (used by ST quite a lot). Caused silent SVD parsing failures before and the Peripherals window saidNo SVD File Loaded
. - Issue #229: Better handling of multiple anonymous unions and structs in the same data structure. Only the first one was shown previously, defect in VSCode really, but had to find a workaround.
- Issue #179: Depending on how the compiler was used, only static variables declared in files in the current directory were being displayed. It was an issue with how
objdump
andgdb
behaved differently. Not a perfect fix. Use Watch Window when in doubt and report any further issues and discrepancies. - Issues with
serialport
module: Updated to work with the latest version of VSCode/Electron. This will be an ongoing problem but hopefully, we can keep up with new releases of VSCode better in the future. When VSCode moves to a new version of Electron this extension has to be updated. For those adventurous enough, there is a script you can use to generate a compatible version yourself.
- Issue #232: SVD Enumerated types
- New Features
- Preliminary support for C++ de-mangled names. In
launch.json
, there is now a configuration option"demangle"
to enable de-mangling of symbols both by GDB and Cortex-Debug. We may remove this property in the future and demangle all the time. All users are encouraged to enable this to see if it affects debugging in a negative way. With C++, there can be a lot of issues related to overloading and templates. Please report issues. - There is a new
launch.json
configuration option"serverArgs"
to add additional command-line arguments when launching any supported gdb-server (like J-Link, ST-LINK, etc.) - Could be classified as a bug-fix. Before, setting static variable values in the Variables Window did not work. Now, it should work as expected.
- There were some performance enhancements done for loading the Variables window when Global or Static scopes were expanded. Noticeable when single-stepping in large executables.
- New setting
flattenAnonymous
which will flatten anonymous structs/unions. Default=false - New setting
registerUseNaturalFormat
which will display registers either in Natural format or Hex: Default=true - The command
View Disassembly (Function)
now supports a regular expression as input. It will try an exact match for a function name first. But, it that fails treats the input string as a regular expression, and if the input string ends with/i
it is treated as case-insensitive. As always, if there are multiple matches, you have to pick one. - You can now specify the
numberOfProcessors
and thetargetProcessor
to debug inlaunch.json
when there are multiple cores/processors in the DAP chain. Cortex-Debug will allocate the required number of TCP ports and use the right one(s) for the processor. This has been tested withpyOCD
andOpenOCD
- Preliminary support for C++ de-mangled names. In
- Fixed an issue where in an attach type debug session, if the disconnect button was used to end debugging, the gdb-server (like OpenOCD) kept running preventing a future attach/launch from starting until the server was killed manually.
- New Features
- Added
postStartSessionCommands
andpostRestartSessionCommands
configuration options tolaunch.json
. These gdb commands are executed after a debug session has (re)started. These options apply to bothlaunch
andattach
sessions. However,postStartSessionCommands
is ignored ifrunToMain
is enabled. See Issue 197. You can use this feature to issue acontinue
or request a gdb-server like OpenOCD to perform a sync, etc. - Added
openOCDPreConfigLaunchCommands
configuration option tolaunch.json
. This is similar toopenOCDLaunchCommands
but are executed before any OpenOCD config files are loaded. - Added the ability to use an expression for the start address of a memory window. This can be any valid expression understood by GDB
- Added
- There are several changes related to RTOS thread aware debugging. In general, the experience is much better but we have become aware that many gdb-servers are not fully compliant or get out of sync with gdb and thus Cortex-Debug. This is especially true at the (re)start of a debug session.
- Better tracking of thread creation/exiting and the entire program exiting
- When the debugger pauses due to any reason, the proper thread is highlighted in the Call Stack Window
- Variables and Watch windows better track the current thread/frame. Same is true for hover and REPL; expressions evaluate in the context of the currently selected thread/frame in the Call Stack Window.
- The gdb-server (like OpenOCD/pyocd) now starts in the same directory as the
cwd
option inlaunch.json
. Before, it was undefined. - Fixed a crash if an SVD peripheral had no registers. Issue #208
- Fixed an issue where setting a variable was not working in the Variables Window. There may still be an issue with file static variables.
NOTE: V0.3.0 has a few backwards incompatible changes that may require you to update your launch.json
file.
- The deprecated launch configuration types (
jlink-gdb
,openocd-gdb
,pyocd-gdb
,stutil-gdb
, andpe-gdb
) have now been removed - all launch configurations for Cortex-Debug should now use thecortex-debug
type. - There are now no SVD files bundled with the main Cortex-Debug extension; these SVD files added significant bulk to the download sizes for the main extension install and update while not being always needed and not changing often. The bundled SVD files will be separated out into separate "Device Support Pack" extensions that target particular microcontrollers (or families of microcontrollers); starting with packs for the STM32F1, STM32F4 and STM32L4 families that had been bundled previously. If you were using your own SVD file specified through the
svdFile
property in yourlaunch.json
then no configuration changes are needed, but if you were using one of the previously auto-detected SVD files through thedevice
property then you will need to install the appropriate "Device Support Packs" (search for "Cortex-Debug" in the extension marketplace).
-
Added support for formatting watch values; add the following format strings:
b
- format in binaryh
orx
- format in hexadecimald
- format in decimalo
- format in octal
These format specifiers are appended to the end of the watch expression separated by a
,
- eg.*(unsigned int *)(0x40011004),b
would display the contents at address0x40011004
in binary. -
Changed core registers to be displayed using their "natural" formatting:
rXX
in decimalsXX
in floating point- stack pointers (
sp
,msp
,psp
) in hexadecimal - program counter (
pc
) in hexadecimal with corresponding symbol location if available - xPSR/cPSR/Control in hexadecimal (this is overridden from the GDB defaults for those registers)
Note that with this change the ability to set formatting for these registers has been disabled; a more flexible formatting solution will be re-added in the future. You can also use a Watch expression for a register to the desired $reg,format to get the format you desire. For instance
$r0,x
will display the register value ofr0
in hexadecimal format. -
Major refactor of the code for the Core Register and Peripheral Register displays; along with bringing (hopefully) improved formatting and UX to this views will make the code much easier to maintain and expand in the future.
-
The SWO grapher view should now be functional again on newer of VSCode. This view is now also theme aware and will adapt the colour scheme for standard elements to work with your theme (you still need to provide appropriate colours for your plots in the graph config).
-
Extension code is now bundled for faster load time and smaller installs
- Added new
servertype
of external for cases where you want to control the GDB server yourself. This could be used for cases where you need to run the GDB server on a different machine, or in cases where there are multiple target cores which may cause the debug server to not operate as expected by cortex-debug. This configuration may require more customizations to the launch.json file than other more automated server types. When thisservertype
is selected a value for thegdbTarget
launch.json property must be supplied. - Added new launch.json options (overrideLaunchCommands, overrideRestartCommands and overrideAttachCommands) to be able to override the default commands run on those operations. In most cases this is not needed, but may be required for
external
server types (by default commands that are compatible with openocd are used forexternal
server types). - Add a
overrideGDBServerStartedRegex
launch.json configuration option - this allows you to provide the system with an alternative regular expression to detect that the GDB server has completed launching and is ready to accept connections. In most cases this will be need - but may be useful in cases where the debug servers output has changed and is no longer recognized. - Major upgrade to the system for finding free ports to use (big thanks to https://github.com/haneefdm for his work on this); should fix recurring problems with port number collisions (e.g. 117(Marus#117)).
- Updates to PyOCD to enable support for CMSIS-Pack specification (added in PyOCD v0.16.0) - thanks to https://github.com/pelrun for this improvement
- Updated watch var name generation to avoid problems with certain expressions containing reserved characters/strings. Should fix issue 159 and partially fix 157.
- Updated PyOCD start detection to work with newer versions - Fixes issue 165
- Updated some embedded SVD files (Thanks https://github.com/clementperon for your PR)
- Fixed parsing of some SVD files (Issue 126 - Thanks https://github.com/mryndzionek for your PR)
- Fixed issues with race condition on start up and improved OpenOCD support; Should fix issues 147, 149 and 150. A huge thanks to https://github.com/haneefdm for this PR, and his ongoing support on the project.
- Ability to specify port ranges used (Thanks https://github.com/blckmn for your PR)
- Fixed issues with serial port source for SWO data (Currently only working on macOS and Linux; Windows support will be restored soon)
- Extension now requires VS Code 1.29 or newer
- Fixed issues with variable substitution
- Fixed issues with blocking run if executable doesn't exist and may be created by the preLaunchTask
- Work around for some issues introduced with VSCode 1.29
- Added initial support for PE Micro Debug Interfaces (Special thanks to https://github.com/danebou for your PR)
- Fixed a number of bugs with the peripheral view - hopefully registers will be updating properly now (note that you can no longer just select a node to expand, you must click on the expand/collapse arrow)
- Fixed issue with people sometimes not being able to set new breakpoints after launching. Special thanks to @xiaoyongdong for his fix and @microwavesafe for his help in testing.
- Fixed issue with setting breakpoints while the target is running
- Fixed issues with the 'Add to Watch' and 'Copy Expression' options in the variable view
- Fixed issues with parsing some Atmel SVD files (Thanks to https://github.com/ivankravets for your PR)
- Allow overriding the armToolchainPath setting on a per lunch configuration basis
- Updated command names for JLink - newer versions of JLink rename the GDB server on Linux and macOS to JLinkGDBServerCLExe - this update searches for both the new JLinkGDBServerCLExe name and, if not found, falls back to the old JLinkGDBServer name.
- Fixed bug with the restart command if the target was currently in a running state.
- Added add a runToMain setting to launch debug requests (not applicable to attach requests).
- Added a searchDir setting for OpenOCD GDB server that allows specifying what directories to search for OpenOCD configuration files. Thanks https://github.com/KaDw
- Improved highlighting in the raw memory view
- Workaround for an issue with enumeratedValue having isDefault instead of a value
- Fixed a bug where it may not detect that a port is in use and get a port conflict when starting the GDB server.
- RTOS Support (configured through the rtos property in your launch.json file)
- Depends on support from GDB Server - currently only J-Link and OpenOCD provide support for RTOS (supported RTOS varies)
- In general if you have RTOS support enabled you should not perform stepping operations before the RTOSs data structures/scheduler have been initialized. Doing so tends to either crash the GDB server or leave it in an inconsistent state which will prevent proper functionality. If you need to debug startup code before the RTOS has been completely initialized then you should disable RTOS support.
- Some basic telemetry has been added
- This telemetry has been added to help me determine what/how features are being used to help me better determine future feature development/improvements.
- No information about your projects source code is collected - only information directly related to the use of cortex-debug is collected. For example the following is collected:
- Number/length of debugging sessions
- Specific features used (peripheral register view, disassembly view, rtos support, memory view, SWO decoding, Graphing, etc.)
- Certain errors within the extension are reported
- GDB Server Used
- Target device (if entered in launch.json)
- Extension Version
- Visual Studio Code Version
- Visual Studio Code Platform
- The information collected is not user-identifiable.
- You can disable all telemetry reporting through the following user/workspace settings:
- setting telemetry.enableTelemetry to false (this will disable telemetry for VS Code and other extensions that respect this setting)
- setting cortex-debug.enableTelemetry to false (this will disable telemetry just for Cortex-Debug)
- Improved support for customizing the launch, attach and restart processes. In most cases these parameters can simply be ignored - but for added flexibility the following settings in your launch.json file can be provided
- preLaunchCommands/preAttachCommands - these are executed near the start of the main launch/attach sequence (immediately after attaching to the target)
- postLaunchCommands/postAttachCommands - these are executed at the end of the main launch/attachSequence
- preRestartCommands - these are executed at the start of the restart sequence (immediately following interrupting the processor)
- postRestartCommands - these are executed at the end of the restart sequence
- Fixes for advanced SWO Decoders
- Workaround for issues with st-util GDB server on Windows environment
- Added ability to select value for matting in the Core and Peripheral Register Views (Right click and Select "Set Value Format")
- Preserve state for Core and Peripheral Register Views (Set format and expanded) from one debug session to the next.
- Syntax highlighting for the raw memory view.
- Enabled setting breakpoints in rust code
- Improved ITM console decoder
- Fixed ITM configuration GDB macros to work properly with rust code
- Fixed issues with parsing dimIndex elements in some SVD files.
- Improved SVD parsing:
- Fields now support bit ranges being defined with and elements; This would have impacted SVD files supplied by Nordic Semiconductor, Fujitsu and Spansion
- Improved support for repeating fields/registers for "array" style repeats, versus explicitly named repeats; This would have impacted SVD files supplied by Nordic Semiconductor, Microchip/Atmel, and some of NXP's LPC line
- Support for register clusters, to group multiple closely related registers, within peripherals; This would have impacted SVD files supplied by Nordic Semiconductor and Microchip/Atmel
- Fixed issue with values being displayed as if they were signed.
- Improved display of Write-Only registers
- Improved behaviour with the Disassembly View:
- Manual triggered disassembly names will now match those automatically generated by missing source/forced disassembly mode - prevents it from opening two copies of the disassembly.
- If there are multiple functions with the same symbol name (two static functions with the same name in two different compilation units) you can now choose between them when manually opening a disassembly view.
- If you are focused on a manual disassembly view for the current frame the debugger will use instruction level stepping, instead of source line level stepping.
- Added a "postLaunchCommands" property to the supported launch.json properties. This should be an array of GDB commands to send after the main launch/attach sequence (you do not need to include things like "target extended-remote ...", "load", or "monitor reset" as these are generated automatically).
- The update has a significant refactoring of code to make supporting the expanding list of GDB Servers more feasible. From the user side this necessitates updating your launch.json files as all debug types have now been combined into one common cortex-debug type
- The typical changes needed are to replace *"type": "-gdb" in your launch.json file with "type": "cortex-debug" and "servertype" : "";
- The extension will attempt to map old configurations automatically - but this may not work in all cases; additionally there launch.json editor will not recognize the old types any more
- You no longer specify paths to the individual tools in your launch.json file; now there are settings you can set (either user level or workspace level) for paths to the individual GDB servers as well as the arm toolchain. For the arm toolchain path the setting should point to the toolchains bin directory - not an individual executable - as multiple tools from the toolchain are now used (current arm-none-eabi-gdb and arm-none-eabi-objdump; but possibly others in the future)
- A globals and static scope has been added to the variables view
- A disassembly view has been added. This can show up in three possible ways:
- You can manually view the disassembly for a particular function by selecting the "Cortex-Debug: View Disassembly (Function) command from the command palette and entering the function name. (While you can view the disassembly in this case, stepping will still be based upon source lines currently)
- If the source file cannot be located it will automatically disassemble and display the current function (In this case stepping is by instruction)
- You can force it to always disassemble through the "Cortex-Debug: Set Force Disassembly" command and selecting the "Forced" option.
- SWO Decoding has been significantly overhauled
- It is now possible to use a serial port (such as a FTDI USB to UART) to capture SWO data, allowing the use of SWO output on probes that do not support it natively or have poor performance. To use this set the "source" key under "swoConfig" to the UART device (COM port on Windows).
- The ITM, DWT and TPIU registers needed to match the configuration in the launch.json file will be set automatically; avoiding the need for your firmware to make the configurations. SWO output will still need to be enabled in your firmware though, as this part of the configuration is microcontroller specific.
- A number of configuration options have changed; please edit your launch.json file
- Initial support for the Black Magic Probe has been added; this server has not been tested extensively yet, so there may still be some issues. SWO output through the probe is not currently support when using the Black Magic Probe.
- Fixed issue with Peripheral Register viewer not working after the first launch request
- Fixed a bug with the variables and watches view incorrectly updating the value on a struct/array when a contained element changed
- Updated the view memory output format to match the format used by the hexdump for VSCode extension (https://marketplace.visualstudio.com/items?itemName=slevesque.vscode-hexdump) - this will enable the syntax highlighting, and hopefully in the future the inspector, from that plugin.
- Added initial support for texane's STLINK utilities st-util GDB server (https://github.com/texane/stlink) - this configuration does not support SWO output.
- Enabled updating registers and fields (Read/Write or Write-Only in the SVD definition) in the Cortex Peripherals view - Right click on the register/field and select "Update"
- Enabled copying registers and fields values in the Cortex Peripherals and Cortex Registers Views - Right click on the register/field and select "Copy Value"
- Fixed possible freeze with memory viewer command and addresses above 0x80000000
- Improved parsing of SVD definitions (registers without fields; repeating registers (dim, dimInteger, dimIncrement))
- Added initial support for PyOCD GDB Server (SWO not supported)
- Initial Public Preview on VS Code Market Place