You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When installing from a lockfile, we need to know the url of each asset for performance and reliability reasons (we don't want to go through the index page again), its size (start the largest download first, progress bars) and its hash (for integrity). In our lockfile this looks as shown below. With the current pypi url structures this means repeating https://files.pythonhosted.org/packages/ for everything single distribution and storing two hex ids, e.g. for the source dist in the example one is the 63/09/c1bc53dab74b1816a00d8d030de5bf98f724c52c1635e07681d312f20be8 in the url and then the f30c3cb33b24454a82faecaf01b19c18562b1e89558fb6c56de4d9118a032fd5 for the hash.
I'm looking for reducing the amount of string we have to store in a lockfile. A smaller lockfile is easier to audit, faster to parse and makes better git diff. For the urls specifically, it would be nice lockfile to have a way to construct a direct url without serializing the entire url, possibly for pypi, ideally even cross index. Let's say e.g. we serialize file-prefix = "https://files.pythonhosted.org/disitribution/" and can then find every file at https://files.pythonhosted.org/disitribution/ + sha256 + / + distribution filename.
Is that something that the warehouse maintainers would be interested in? I know fairly few about the overall architecture of warehouse, so i'd be interested in what you think about this, if there are any technical constraints from the CDN or other considerations to be aware of.
When installing from a lockfile, we need to know the url of each asset for performance and reliability reasons (we don't want to go through the index page again), its size (start the largest download first, progress bars) and its hash (for integrity). In our lockfile this looks as shown below. With the current pypi url structures this means repeating
https://files.pythonhosted.org/packages/
for everything single distribution and storing two hex ids, e.g. for the source dist in the example one is the63/09/c1bc53dab74b1816a00d8d030de5bf98f724c52c1635e07681d312f20be8
in the url and then thef30c3cb33b24454a82faecaf01b19c18562b1e89558fb6c56de4d9118a032fd5
for the hash.I'm looking for reducing the amount of string we have to store in a lockfile. A smaller lockfile is easier to audit, faster to parse and makes better git diff. For the urls specifically, it would be nice lockfile to have a way to construct a direct url without serializing the entire url, possibly for pypi, ideally even cross index. Let's say e.g. we serialize
file-prefix = "https://files.pythonhosted.org/disitribution/"
and can then find every file athttps://files.pythonhosted.org/disitribution/
+ sha256 +/
+ distribution filename.Is that something that the warehouse maintainers would be interested in? I know fairly few about the overall architecture of warehouse, so i'd be interested in what you think about this, if there are any technical constraints from the CDN or other considerations to be aware of.
If we could construct the url, we could slim this entry:
The text was updated successfully, but these errors were encountered: