Unable to omit unwind tables from binary when using ReleaseSmall + LLVM

Zig Version

0.14.0-dev.2164+6b2c8fc68

Steps to Reproduce and Observed Behavior

$ cat x.zig
pub fn main() void {}
$ zig build-exe x.zig -fno-unwind-tables -OReleaseSmall
error: <inline asm>:1:3: this directive must appear between .cfi_startproc and .cfi_endproc directives
         .cfi_undefined %rip
         ^

$

Expected Behavior

Successful compilation.

[Rexicon226]

This is the line of inline asm that's causing this:

zig/lib/std/start.zig

Line 250 in e5f5229

\\ .cfi_undefined %%rip

added in f289b82

[alexrp]

I have an upcoming PR that improves the compiler's handling of unwind tables, including adding a builtin.unwind_tables decl. With that, we can more easily decide whether to emit the .cfi_undefined directive.

Similar fix is needed in clone().

[alexrp]

#22035 is the first step here. We'll need a zig1.wasm update before we can actually start using @import("builtin").unwind_tables in standard library code, though.

[alexrp]

#22035 has been merged so this can fixed once the next zig1.wasm update happens.

Sorry, what's the problem with making this issue a reason to do a zig1.wasm update and... doing it? AFAIK all it takes is a trustable contributor (i.e. core team member) to commit the new binary. Would doing this break someone's workflow or involve any highly expensive computations either for the CI or anything else or is there any other reason to postpone this? (Just curious)

[Rexicon226]

zig1.wasm updates bloat the binary because blobs are stored in whole, not as diffs in git repos. the core team tries to limit how many updates are done as much as possible. it also makes it easier to maintainers in the future to bootstrap up from stage1.

[alexrp]

There's a zig1.wasm update in #22386 so it's not too far off now. One way or another, it will happen before 0.14.0.

[jacobly0]

Why not use @hasDecl to add support before a zig1 update, then the feature is only missing from zig1 itself?

[alexrp]

Working on this now since a zig1.wasm update is in.

Why not use @hasDecl to add support before a zig1 update, then the feature is only missing from zig1 itself?

Could have done, I just didn't feel like going back later to remove the @hasDecls.