Format function arg defaults with the constants-as-written printer by conrade-ctc · Pull Request #1069 · compiler-research/CppInterOp

conrade-ctc · 2026-07-16T16:24:53Z

GetFunctionArgDefault re-printed default arguments at maximum float precision, so 3.14 rendered as 3.1400000000000001. Print them with printPretty, the interpreter's ASTContext, and ConstantsAsWritten = true instead: literal leaves keep their source spelling (4.0 stays 4.0).

Two clang-side gaps remain: the maximum-precision expansion where ConstantsAsWritten cannot apply (fix under review in llvm/llvm-project#218471), and cooked UDL floats (90.0_deg prints as 90._deg). A canary test expects the old output before clang 24 and the fixed output from clang 24 on.

codecov · 2026-07-16T16:29:58Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.70%. Comparing base (1795887) to head (5c372a8).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1069      +/-   ##
==========================================
+ Coverage   87.68%   87.70%   +0.01%     
==========================================
  Files          23       23              
  Lines        6382     6391       +9     
==========================================
+ Hits         5596     5605       +9     
  Misses        786      786

Files with missing lines	Coverage Δ
lib/CppInterOp/CppInterOp.cpp	`90.51% <100.00%> (+0.02%)`	⬆️

Files with missing lines	Coverage Δ
lib/CppInterOp/CppInterOp.cpp	`90.51% <100.00%> (+0.02%)`	⬆️

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

vgvassilev · 2026-07-16T16:33:31Z

+      // is not a plain literal.
+      errno = 0;
+      char* ParseEnd = nullptr;
+      double DefaultArgValue = std::strtod(Result.c_str(), &ParseEnd);


Maybe we can take a step back here -- is that the way clang does this?

good point... claude pointed out to me that there is a more standard, generic fix... two possibilities... let me try that them, and repost... might close one of the two PRs actually, since they are related. Give me a sec...

Good call — reworked. It now does what SemaCodeComplete's GetDefaultValueString does: read the parameter's getDefaultArgRange() source text via Lexer::getSourceText (including stripping the sometimes-present leading =, per the FIXME there), falling back to the AST pretty-print only when there's no usable range. That makes the output as-written by construction, so the float-precision FIXME and the std::stod normalization are gone entirely. Literal defaults now render exactly as written (4.0 where the old code produced 4.) — existing test expectations updated to match.

github-actions · 2026-07-16T16:42:48Z

clang-tidy review says "All clean, LGTM! 👍"

github-actions · 2026-07-16T17:58:54Z

clang-tidy review says "All clean, LGTM! 👍"

vgvassilev · 2026-08-22T06:47:29Z

+    // Render the default as written: read its source range, the way clang's
+    // own code completion formats default arguments (GetDefaultValueString
+    // in SemaCodeComplete). Pretty-printing the AST instead reproduces
+    // floating literals at representation precision ("3.1400000000000001"


I believe that's a bug in clang -- it should print the value as written. Can we follow up there, and land that as a temporary workaround?

makes sense, i'll iterate on that and reproducing the bug upstream. i'll post a PR for that separately, and we can treat this as a work-around. I'll update this PR as well with the proper guards so that we can clean up later.

also, in the process of iterating on the upstream, a lighter version of this diff that uses a different pathway was found, so i'm verifying it and probably will update this PR to swap in that change instead.

Reworked: the patch now prints defaults with printPretty, the interpreter's ASTContext, and ConstantsAsWritten = true, instead of the hand-rolled source-text read. All existing tests pass unchanged. The clang-side precision bug is being fixed in llvm/llvm-project#218471; until then, defaults that ConstantsAsWritten cannot cover (cooked UDL floats, invalid-range literals) still print at maximum precision. A version-gated canary test expects the old output before clang 24 and the fixed output from clang 24 on, so a cherry-pick or a slipped landing shows up as a named failure.

vgvassilev · 2026-08-24T18:26:46Z

-    }
+    ASTContext& Ctx = getASTContext();
+    PrintingPolicy Policy(Ctx.getLangOpts());
+    Policy.ConstantsAsWritten = true;


Oh, that’s nice. And probably kills our upstream story probably for good.

@vgvassilev, do you think the UDL hole that is listed above is worth a PR to clang as well? it's a separate gap that might be valuable to clean up.

Oh, ok. Did not realize that’s a disk access. That’s pretty slow for the pch case and requires headers to be around..

hmmm, by UDL i mean user-defined-literal, not the disk one :) Just want to make sure we're on the same page... there is apparently a minor hole in the UDL float impl that mixes 90.0_deg and 90._deg for example... claude identified that as a follow up that is related and needs a separate treatment upstream.

github-actions

clang-tidy made some suggestions

github-actions · 2026-08-24T20:44:14Z

clang-tidy review says "All clean, LGTM! 👍"

Print the default-argument expression with printPretty, the interpreter's ASTContext, and ConstantsAsWritten = true. The policy prints literal leaves from their source text, so defaults keep their spelling (4.0 stays "4.0"). This replaces the AST re-print at maximum float precision, which turned 3.14 into 3.1400000000000001. Co-developed-with-the-help-of: Claude Code (Fable 5, human in the loop)

Fails on the first LLVM that prints shortest round-trip floats (llvm/llvm-project#218471), which changes the output for defaults the constants-as-written policy cannot cover. Co-developed-with-the-help-of: Claude Code (Fable 5, human in the loop)

Print the default-argument expression with printPretty, the interpreter's ASTContext, and ConstantsAsWritten = true. The policy prints literal leaves from their source text, so defaults keep their spelling (4.0 stays "4.0"). This replaces the AST re-print at maximum float precision, which turned 3.14 into 3.1400000000000001. A canary test fails on the first LLVM that prints shortest round-trip floats (llvm/llvm-project#218471), which changes the output for defaults the policy cannot cover. Content matches compiler-research#1069. Co-developed-with-the-help-of: Claude Code (Fable 5, human in the loop)

vgvassilev reviewed Jul 16, 2026

View reviewed changes

conrade-ctc changed the title ~~Keep symbolic floating defaults in GetFunctionArgDefault~~ Format function arg defaults from their source text Jul 16, 2026

conrade-ctc force-pushed the argdefault-symbolic-floating-defaults branch from 189b6aa to 2d07020 Compare July 16, 2026 17:33

conrade-ctc mentioned this pull request Jul 28, 2026

Emit discardable-ODR variables via odr-use instead of UsedAttr in GetVariableOffset #1068

Merged

conrade-ctc force-pushed the argdefault-symbolic-floating-defaults branch from 2d07020 to 1bc99c8 Compare August 21, 2026 21:05

vgvassilev reviewed Aug 22, 2026

View reviewed changes

conrade-ctc mentioned this pull request Aug 24, 2026

[clang][APFloat] Print floating-point literals in the shortest round-trip form llvm/llvm-project#218471

Open

conrade-ctc force-pushed the argdefault-symbolic-floating-defaults branch 2 times, most recently from 2535e74 to 1c98a0f Compare August 24, 2026 18:25

vgvassilev reviewed Aug 24, 2026

View reviewed changes

conrade-ctc changed the title ~~Format function arg defaults from their source text~~ Format function arg defaults with the constants-as-written printer Aug 24, 2026

github-actions Bot reviewed Aug 24, 2026

View reviewed changes

Comment thread unittests/CppInterOp/FunctionReflectionTest.cpp

Comment thread unittests/CppInterOp/FunctionReflectionTest.cpp

conrade-ctc force-pushed the argdefault-symbolic-floating-defaults branch from 1c98a0f to e900344 Compare August 24, 2026 20:28

Emery Conrad added 2 commits August 27, 2026 09:24

conrade-ctc force-pushed the argdefault-symbolic-floating-defaults branch from e900344 to 5c372a8 Compare August 27, 2026 14:24

Uh oh!

Conversation

conrade-ctc commented Jul 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Jul 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Choose a reason for hiding this comment

Uh oh!

conrade-ctc Jul 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Jul 16, 2026

Uh oh!

github-actions Bot commented Jul 16, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Aug 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

conrade-ctc commented Jul 16, 2026 •

edited

Loading

codecov Bot commented Jul 16, 2026 •

edited

Loading

conrade-ctc Jul 16, 2026 •

edited

Loading