Fix ICE caused by invalid spans for shrink_file

[chenyukang]

Fixes #148732

There are two issues in this function:

1. the original issue is caused by a typo error, which is fixed in the first commit
2. another different ice(Patch span 7..7 is beyond the end of buffer 0) will be reported after fixing the first one, is caused by spans cross file boundaries due to macro expansion. It is fixed in the second commit.

r? @nnethercote

edited: also fixes #148684, added a new testcase for it in the last commit.

[rustbot]

rustc_errors::annotate_snippet_emitter_writer was changed

cc @Muscraft

[Kivooeo]

I have worries that this will not pass aarch64-msvc-1 the same way it was in my PR

cc: #general > Windows path normalization in tests

[chenyukang]

ah, it do have the same issue.

[chenyukang]

so, I read the discussion on zulip, seems there is another standalone bug. so is it a temprary way we skip Windows for this testcase(and add it back after that bug is fixed)?

[Muscraft]

These warnings are coming from this call to is_different. The reason this is happening with AnnotateSnippetEmitter and not HumanEmitter is because HumanEmitter checks for source code availability before it calls is_different. Changing lines 353 to 368 in annotate_snippet_emitter_writer.rs to the following fixes the issue:

let lo = subst.parts.iter().map(|part| part.span.lo()).min()?;
let lo_file = sm.lookup_source_file(lo);
let hi = subst.parts.iter().map(|part| part.span.hi()).max()?;
let hi_file = sm.lookup_source_file(hi);

// The different spans might belong to different contexts, if so ignore suggestion.
if lo_file.stable_id != hi_file.stable_id {
    return None;
}

// We can't splice anything if the source is unavailable.
if !sm.ensure_source_file_source_present(&lo_file) {
    return None;
}

// Account for cases where we are suggesting the same code that's already
// there. This shouldn't happen often, but in some cases for multipart
// suggestions it's much easier to handle it here than in the origin.
subst.parts.retain(|p| is_different(sm, &p.snippet, p.span));

if subst.parts.is_empty() { None } else { Some(subst) }

[chenyukang]

yes, this can fix the issue.
I have a question, the old code invoke should_show_source_code to check with ignored_directories_in_source_blocks, we don't need this anymore?

[chenyukang]

updated the code and tests.
@Kivooeo i'm not sure whether your PR could be unblocked by this commit, you can have a try.

[Kivooeo]

Thanks, I'll try later today

[Muscraft]

yes, this can fix the issue. I have a question, the old code invoke should_show_source_code to check with ignored_directories_in_source_blocks, we don't need this anymore?

I switched it so that it would match splice_lines, which is where I got the solution from. Going back to using should_show_source_code to maintain consistency with the old code would be a perfectly reasonable thing to do.

[Kivooeo]

@bors try jobs=aarch64-msvc-1

[rust-bors]

💔 Test for eca91bd failed: CI. Failed jobs:

• try - aarch64-msvc-1 (web logs, extended logs)

[nnethercote]

r=me once the comments are addressed.

@rustbot author

[chenyukang]

@bors r=nnethercote

[bors]

📌 Commit 4a2c9a1 has been approved by nnethercote

It is now in the queue for this repository.

[bors]

⌛ Testing commit 4a2c9a1 with merge 5dbf406...

[Zalathar]

Bors no, this is already merged.

@bors r- retry