stacktraces: follow-up to "Enable method signature printing for inlined frames" by mlechu · Pull Request #61699 · JuliaLang/julia

mlechu · 2026-04-30T21:07:47Z

Two changes to #53925:

When recursing on linetable in append_lineinfo, we ended up with
the wrong pc in the column field, since we were using the innermost
index rather than the one usable with a CodeInstance's .debuginfo.
Store the optimized pc in the "innermost" frame, since DILocation
expects inlinee locations to change more quickly than inliners (see
performance discussion below). Thanks Cody and Jameson for this
suggestion.
Update stacktraces.jl to reconstruct the list of inlined frames upon
seeing a debuginfo, pc pair. See comment; I don't like having both
this code and the old fallback lookup implementation, but it seems
less brittle to just re-create the inlined frame stack than to create it
and match it with the linear frames array from
jl_lookup_code_address.

aviatesk

Thanks, the point 2 is fixing a real bug, not an intentional design in #53925. I missed the case where the outer debuginfo has a linetable indirection.

The existing inlined_test just happens to pass by coincidence as every IR PC of f points at to=1, so the column lands on the right edge regardless.
I crafted this reproducer of the bug that surfaces when the caller has several non-inlined statements before an inlined call:

@inline f_inner3(x) = x > 0 ? x : error("neg: $x")
function f_parent3(a, b, c, d)
    s = 0
    @noinline begin   # keep `+` as invokes so codelocs has `to=0` entries
        s += a; s += b; s += c; s += d
    end
    return f_inner3(s)
end
let st = try f_parent3(1, 2, 3, -10) catch; stacktrace(catch_backtrace()) end
    sf = only(filter(sf -> sf.func === :f_inner3 && sf.inlined, st))
    @test sf.linfo isa Core.MethodInstance
    @test sf.linfo.def === which(f_inner3, (Int,))
end

This fails on master, but passes with your patch.

Point 1 also LGTM, so ready to come out of draft from my side.

topolarity · 2026-05-01T17:07:57Z

    if (line != -1) {
        if (pc > 0)
-            jl_safe_fprintf(s, "%s at %s:%d:%d%s\n", func_name, file_name, line, pc, inlined_str);
+            jl_safe_fprintf(s, "%s at %s:%d (pc: %d)%s\n", func_name, file_name, line, pc, inlined_str);


I think our frame construction fills in .pc even for C frames, so that may also need follow-up.

topolarity

Great find. Probably means that we need better testing infra. for the various DebugInfo variants long-term.

framepc may be a slightly better name than origpc

Thanks for the fix-up @mlechu !

mlechu · 2026-05-01T20:58:34Z

Failure (OOM in LLVM on 32-bit windows) looks real; the gnuprofiling build shows section .debug_info in sys.so has gotten 3x larger

mlechu · 2026-05-01T23:30:56Z

I suspect the cause is storing way more unique locations given that optimized IR usually has more unique indices. For my current use case (byte provenance) it doesn't matter if we use the innermost-linetable indices, but I'm not sure if that's OK for the purposes of the original change.

Change / approach is conceptually right, but size regression needs to be resolved before merging

…tion (#61817) This change adds a basic C API for reading `jl_debuginfo_t`, both in its current form and with byte-precise provenance to be generated by JuliaLowering. ``` (debuginfo, pc) -> (firstbyte, lastbyte) (debuginfo, byte) -> (line, column) (debuginfo, pc) -> (firstline, firstcolumn) debuginfo -> first_line_of_entire_codeinfo ``` Note that the byte-precise provenance is not yet produced (and I've marked those branches unreachable). I'm making this change first so I can use them to try and resolve issues in #61699 and #53925. (Namely, I'm going to try and compute inlined frames from julia debuginfo rather than DWARF.) ## Implementation of byte-precise info storage The API shouldn't depend on how we do this, but I'll describe it here for reference. The simplicity of our current `Core.DebugInfo` (ref. #52415) is thanks to simple line numbers looking the same as indices into the next debuginfo, so they can be stored in the same way. Unfortunately, byte-precise information breaks this: from an index, we want to be able to know two byte positions, what lines they fall on, and how far from the beginning of those lines they are. I intend to store this info by putting a string in `linetable` like we do for `di.codelocs`. The compression format is described in julia.h, but only a couple points are relevant here: - Like `di.codelocs` today, I made a reasonable attempt to compress it (so we don't blow up the sysimg) without trying to be too clever. - Position lookup for a single IR statement can still be done without decompressing everything. I've already implemented the compression/decompression code (in JuliaLowering), but it's not part of this PR. This format might be torn out when we start trying to store AST provenance. --------- Co-authored-by: Cody Tapscott <topolarity@tapscott.me>

…tion (JuliaLang#61817) This change adds a basic C API for reading `jl_debuginfo_t`, both in its current form and with byte-precise provenance to be generated by JuliaLowering. ``` (debuginfo, pc) -> (firstbyte, lastbyte) (debuginfo, byte) -> (line, column) (debuginfo, pc) -> (firstline, firstcolumn) debuginfo -> first_line_of_entire_codeinfo ``` Note that the byte-precise provenance is not yet produced (and I've marked those branches unreachable). I'm making this change first so I can use them to try and resolve issues in JuliaLang#61699 and JuliaLang#53925. (Namely, I'm going to try and compute inlined frames from julia debuginfo rather than DWARF.) ## Implementation of byte-precise info storage The API shouldn't depend on how we do this, but I'll describe it here for reference. The simplicity of our current `Core.DebugInfo` (ref. JuliaLang#52415) is thanks to simple line numbers looking the same as indices into the next debuginfo, so they can be stored in the same way. Unfortunately, byte-precise information breaks this: from an index, we want to be able to know two byte positions, what lines they fall on, and how far from the beginning of those lines they are. I intend to store this info by putting a string in `linetable` like we do for `di.codelocs`. The compression format is described in julia.h, but only a couple points are relevant here: - Like `di.codelocs` today, I made a reasonable attempt to compress it (so we don't blow up the sysimg) without trying to be too clever. - Position lookup for a single IR statement can still be done without decompressing everything. I've already implemented the compression/decompression code (in JuliaLowering), but it's not part of this PR. This format might be torn out when we start trying to store AST provenance. --------- Co-authored-by: Cody Tapscott <topolarity@tapscott.me>

mlechu · 2026-05-28T19:33:46Z

Approach and PR message updated; this is ready for re-review.

…61939) Fix #33457. Nightly: ``` julia> begin # line 2 [y for _ in 1:10] # line 4 end ERROR: UndefVarError: `y` not defined in `Main` Suggestion: check for spelling errors or missing imports. Stacktrace: [1] #2 @ ./none:-1 [inlined] [2] iterate @ ./generator.jl:48 [inlined] [3] collect(itr::Base.Generator{UnitRange{Int64}, var"#2#3"}) @ Base ./array.jl:833 [4] top-level scope @ REPL[1]:3 ``` This change: ``` ERROR: UndefVarError: `y` not defined in `Main` Suggestion: check for spelling errors or missing imports. Stacktrace: [1] #12 @ ./REPL[1]:3 [inlined] [2] iterate @ ./generator.jl:48 [inlined] [3] collect(itr::Base.Generator{UnitRange{Int64}, var"#12#13"}) @ Base ./array.jl:833 [4] top-level scope @ REPL[1]:3 ``` As a sneak peek, this is further improved with #61699: ``` ERROR: UndefVarError: `y` not defined in `Main` Suggestion: check for spelling errors or missing imports. Stacktrace: [1] (::var"#12#13")(::Int64) @ Main REPL[1]:3 [inlined] [2] iterate(::Base.Generator{UnitRange{Int64}, var"#12#13"}) @ Base generator.jl:48 [inlined] [3] collect(itr::Base.Generator{UnitRange{Int64}, var"#12#13"}) @ Base array.jl:833 [4] top-level scope @ REPL[1]:3 ``` I doubt this frame is important (it being a lambda is considered an implementation detail IIRC) but (1) we shouldn't show users a nonsense frame, and (2) this fixes another case where we produce malformed DebugInfo.

topolarity

How exciting this worked out in the end!

I've opened JuliaLang/llvm-project#52 to expand the DWARF column to 32-bit, but I think we can land this as-is

Co-authored-by: Shuhei Kadowaki <aviatesk@gmail.com>

Put PC in most-inlined frame

We want to decode `DebugInfo` rather than using linfo from `jl_lookup_code_address`. This temporarily matches flisp.

We no longer get the dummy `(::Tuple)() at <missing>:0` in the stacktrace since we produced the opaque closure by hand with abnormal debuginfo. Seeing no better alternative, still inspect each of the frames to check we don't crash, but the absence of the frame means this test probably can't hit the original crash anyway.

Avoid construction using Core.OpaqueClosure

aviatesk · 2026-06-02T07:48:33Z

Thanks again for fixing this issue and getting it merged. I really appreciate it!

mlechu requested a review from aviatesk April 30, 2026 21:07

mlechu mentioned this pull request May 1, 2026

Debuginfo: misc. cleanup #61701

Merged

aviatesk reviewed May 1, 2026

View reviewed changes

mlechu force-pushed the debuginfo-pc-fix branch from ccc7603 to 34b44e4 Compare May 1, 2026 16:20

mlechu marked this pull request as ready for review May 1, 2026 16:20

topolarity reviewed May 1, 2026

View reviewed changes

topolarity previously approved these changes May 1, 2026

View reviewed changes

This was referenced May 15, 2026

Core.DebugInfo: expand C API, mostly to handle byte-precise information #61817

Merged

[WIP] Recover inlined frames using Core.Debuginfo instead of DWARF #61819

Closed

stacktraces: Remove MI printing for inlined frames #61866

Closed

mlechu closed this May 20, 2026

mlechu reopened this May 20, 2026

mlechu force-pushed the debuginfo-pc-fix branch 2 times, most recently from de72d43 to 95ac8f5 Compare May 21, 2026 16:58

mlechu force-pushed the debuginfo-pc-fix branch from 95ac8f5 to b119158 Compare May 22, 2026 20:04

mlechu force-pushed the debuginfo-pc-fix branch 6 times, most recently from cd05077 to ab6a0c2 Compare May 28, 2026 19:21

mlechu mentioned this pull request May 29, 2026

Fix none:-1 stack frame from generator/comprehension expressions #61939

Merged

topolarity mentioned this pull request May 29, 2026

DILocation: Support 32-bit column number JuliaLang/llvm-project#52

Open

topolarity reviewed May 29, 2026

View reviewed changes

Comment thread src/codegen.cpp

topolarity approved these changes May 29, 2026

View reviewed changes

Comment thread base/stacktraces.jl

topolarity reviewed May 29, 2026

View reviewed changes

Comment thread base/stacktraces.jl

mlechu and others added 10 commits June 1, 2026 09:59

update StackFrame comment: pc holds a real column when from_c

50b2399

Add test

d3e0275

Co-authored-by: Shuhei Kadowaki <aviatesk@gmail.com>

Use outer pc, not index into innermost .linetable

c852887

Put PC in most-inlined frame

Update stacktraces.jl to find MIs

724f15b

Fix bad ccall

2b164ba

Weaken a JL stacktrace test

0268b16

We want to decode `DebugInfo` rather than using linfo from `jl_lookup_code_address`. This temporarily matches flisp.

Move and strengthen OC stacktrace tests

6c38e47

Avoid construction using Core.OpaqueClosure

tweaks

1b51e97

make assert message more helpful

85e3619

mlechu force-pushed the debuginfo-pc-fix branch from ab6a0c2 to 85e3619 Compare June 1, 2026 16:59

mlechu added the merge me PR is reviewed. Merge when all tests are passing label Jun 1, 2026

topolarity merged commit 39bd0b4 into JuliaLang:master Jun 1, 2026
10 of 11 checks passed

topolarity removed the merge me PR is reviewed. Merge when all tests are passing label Jun 1, 2026

topolarity mentioned this pull request Jun 3, 2026

[debug] Support 32-bit column number in DILocation llvm/llvm-project#201269

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

stacktraces: follow-up to "Enable method signature printing for inlined frames"#61699

stacktraces: follow-up to "Enable method signature printing for inlined frames"#61699
topolarity merged 10 commits into
JuliaLang:masterfrom
mlechu:debuginfo-pc-fix

mlechu commented Apr 30, 2026 •

edited

Loading

Uh oh!

aviatesk left a comment

Uh oh!

topolarity May 1, 2026

Uh oh!

topolarity left a comment

Uh oh!

mlechu commented May 1, 2026

Uh oh!

mlechu commented May 1, 2026

Uh oh!

mlechu commented May 28, 2026

Uh oh!

Uh oh!

topolarity left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

aviatesk commented Jun 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Uh oh!

Conversation

mlechu commented Apr 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

aviatesk left a comment

Choose a reason for hiding this comment

Uh oh!

topolarity May 1, 2026

Choose a reason for hiding this comment

Uh oh!

topolarity left a comment

Choose a reason for hiding this comment

Uh oh!

mlechu commented May 1, 2026

Uh oh!

mlechu commented May 1, 2026

Uh oh!

mlechu commented May 28, 2026

Uh oh!

Uh oh!

topolarity left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

aviatesk commented Jun 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

mlechu commented Apr 30, 2026 •

edited

Loading