Fix fallback logic in analysis.getPositionContext

[llogick]

Most lines are indented, ie line[0] == ' '(a space).

Tokenizing line as the sole buffer returns location offsets relative/within to that line, but are expected to be indexes within handle.tree.source .

[codecov]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (6478f1a) 76.58% compared to head (b4d49a9) 76.59%.

Files	Patch %	Lines
src/analysis.zig	90.00%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1754   +/-   ##
=======================================
  Coverage   76.58%   76.59%           
=======================================
  Files          35       35           
  Lines       10088    10090    +2     
=======================================
+ Hits         7726     7728    +2     
  Misses       2362     2362           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Techatrix]

This looks like something that could be covered by a test case in position_context.zig.

[llogick]

Having done this makes me realize having that logic is very dicey and

zls/src/features/completions.zig

Line 1338 in 05ae41e

handle.tree.tokens.items(.start)[el_dot_context.identifier_token_index],

should just use tok_loc.end or

zls/src/features/completions.zig

Line 1197 in 05ae41e

switch (try Analyser.getPositionContext(arena, handle.tree.source, source_index, false)) {

look_ahead = true here (haven't tested for side effects).

[llogick]

This is a low priority PR, the test changes are worth having, but in my opinion the fallback logic shouldn't exist (and obviously nothing depends on it currently).