Revise `alltypes.h` update process. NFC #24773

kleisauke · 2025-07-23T13:17:26Z

Align it with musl, i.e. only update alltypes.h.in and regenerate alltypes.h using a script.

Align it with musl, i.e. only update `alltypes.h.in` and regenerate `alltypes.h` using a script.

…ltypes.h.in

sbc100

Great!

sbc100 · 2025-07-23T15:11:17Z

system/lib/libc/musl/arch/emscripten/bits/update_alltypes.sh

+
+musl_srcdir="$PWD/../../.."
+musl_includedir="$musl_srcdir/include"
+sed -f $musl_srcdir/tools/mkalltypes.sed ./alltypes.h.in $musl_includedir/alltypes.h.in > alltypes.h


Can you put this up in system/lib with the other scripts we have? (or maybe in system/lib/libc/musl?)

sbc100 · 2025-07-23T15:12:08Z

system/lib/libc/musl/arch/emscripten/bits/alltypes.h.in


+#if defined(__FLT_EVAL_METHOD__) && __FLT_EVAL_METHOD__ == 0


Do we need this if/else block here? I don't see it many of the upstream .in files so I think maybe we don't need it?

It appears to be defined only for 32-bit targets in musl. But you're right, this conditional statement is probably unnecessary for Wasm. Commit 4a3bcba resolves this.

sbc100 · 2025-07-23T15:14:07Z

system/lib/libc/musl/arch/emscripten/bits/alltypes.h.in


 #ifndef __cplusplus
-TYPEDEF unsigned wchar_t;
+#ifdef __WCHAR_TYPE__


We can just assume WCHAR_TYPE is present I think.

Ah, it seems the __cplusplus exclusion was still needed. ecefbbd

sbc100 · 2025-07-23T15:16:03Z

system/lib/libc/musl/arch/emscripten/bits/alltypes.h.in

 #endif
-TYPEDEF unsigned wint_t;
+#endif
+TYPEDEF __WINT_TYPE__ wint_t;


This is not needed I think.

This was previously defined here:

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h

Lines 52 to 55 in 90250aa

#if defined(__NEED_wint_t) && !defined(__DEFINED_wint_t)

typedef __WINT_TYPE__ wint_t;

#define __DEFINED_wint_t

#endif

WIntType in LLVM seems to default to SignedInt, see e.g.:
https://github.com/llvm/llvm-project/blob/llvmorg-20.1.8/clang/lib/Basic/TargetInfo.cpp#L131

So, this ensures it remains signed, otherwise, it defaults to unsigned.

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h

Lines 278 to 281 in 4bb13ba

#if defined(__NEED_wint_t) && !defined(__DEFINED_wint_t)

typedef unsigned wint_t;

#define __DEFINED_wint_t

#endif

(commit 4554600 might also be relevant here)

sbc100 · 2025-07-24T20:51:17Z

system/lib/libc/musl/arch/emscripten/bits/alltypes.h

 #define __DEFINED_blksize_t
 #endif

 #if defined(__NEED_blkcnt_t) && !defined(__DEFINED_blkcnt_t)
-typedef int blkcnt_t;
+typedef _Int64 blkcnt_t;


Some of these look like real ABI changes. Perhaps we should try to make this changes actually NFC and do any ABI changes as followups? (or land them first).

This is a bit tricky to review, but blkcnt_t should still be 32-bit, see:

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h

Lines 25 to 29 in 82662c0

// XXX EMSCRIPTEN: ensure it's always 32-bits even in wasm64

#if defined(__NEED_blkcnt_t) && !defined(__DEFINED_blkcnt_t)

typedef int blkcnt_t;

#define __DEFINED_blkcnt_t

#endif

These 32-bit type overrides originate from alltypes.h.in:

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h.in

Lines 17 to 23 in 82662c0

// XXX EMSCRIPTEN: ensure it's always 32-bits even in wasm64

TYPEDEF int blkcnt_t;

TYPEDEF int blksize_t;

TYPEDEF int clock_t;

TYPEDEF unsigned int dev_t;

TYPEDEF int suseconds_t;

TYPEDEF unsigned int wctype_t;

(and alltypes.h is generated from this file)

The musl default definitions starts at line 95:

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h

Line 95 in 82662c0

#define __LITTLE_ENDIAN 1234

Also, other.test_gen_struct_info still passes on this branch, any ABI break would likely have caused it to fail.

With commit 6809780 and the following local change:

--- a/tools/system_libs.py +++ b/tools/system_libs.py @@ -62,7 +62,7 @@ def glob_in_path(path, glob_pattern, excludes=()): def get_base_cflags(build_dir, force_object_files=False, preprocess=True): # Always build system libraries with debug information. Non-debug builds # will ignore this at link time because we link with `-strip-debug`. - flags = ['-g', '-sSTRICT', '-Werror'] + flags = ['-sSTRICT', '-Werror'] if settings.LTO and not force_object_files: flags += ['-flto=' + settings.LTO] if settings.RELOCATABLE:

I now see:

$ ./embuilder.py build sysroot libc{,-mt} --force $ ./embuilder.py build sysroot libc{,-mt} --force --wasm64 $ sha256sum cache/sysroot/lib/wasm{32,64}-emscripten/libc{,-mt}.a dd54801051079bd19040b4d826ded15f7352be10af1ad70d7aa74ce3befb6239 cache/sysroot/lib/wasm32-emscripten/libc.a 6fafd19332c11860b2365efec265fc037080ec97d1a84fafcca9ff338461bd70 cache/sysroot/lib/wasm32-emscripten/libc-mt.a 0310ba126aa4ca060538b6203105b71bfb115c24bcf82d6412c50e0fe041b74c cache/sysroot/lib/wasm64-emscripten/libc.a 66ff6308831fb3df57cc884c10207f73135f4399920e871e83a85f9ee1ac9ab0 cache/sysroot/lib/wasm64-emscripten/libc-mt.a $ git checkout revise-musl-alltypes-h $ ./embuilder.py build sysroot libc{,-mt} --force $ ./embuilder.py build sysroot libc{,-mt} --force --wasm64 $ sha256sum cache/sysroot/lib/wasm{32,64}-emscripten/libc{,-mt}.a dd54801051079bd19040b4d826ded15f7352be10af1ad70d7aa74ce3befb6239 cache/sysroot/lib/wasm32-emscripten/libc.a 6fafd19332c11860b2365efec265fc037080ec97d1a84fafcca9ff338461bd70 cache/sysroot/lib/wasm32-emscripten/libc-mt.a 0310ba126aa4ca060538b6203105b71bfb115c24bcf82d6412c50e0fe041b74c cache/sysroot/lib/wasm64-emscripten/libc.a 66ff6308831fb3df57cc884c10207f73135f4399920e871e83a85f9ee1ac9ab0 cache/sysroot/lib/wasm64-emscripten/libc-mt.a

So, when built without debug information, the libc static libraries are byte-for-byte identical to those built from the main branch.

This is a bit tricky to review, but blkcnt_t should still be 32-bit, see:

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h

Lines 25 to 29 in 82662c0

// XXX EMSCRIPTEN: ensure it's always 32-bits even in wasm64

#if defined(__NEED_blkcnt_t) && !defined(__DEFINED_blkcnt_t)

typedef int blkcnt_t;

#define __DEFINED_blkcnt_t

#endif

These 32-bit type overrides originate from alltypes.h.in:

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h.in

Lines 17 to 23 in 82662c0

// XXX EMSCRIPTEN: ensure it's always 32-bits even in wasm64

TYPEDEF int blkcnt_t;

TYPEDEF int blksize_t;

TYPEDEF int clock_t;

TYPEDEF unsigned int dev_t;

TYPEDEF int suseconds_t;

TYPEDEF unsigned int wctype_t;

(and alltypes.h is generated from this file)
The musl default definitions starts at line 95:

emscripten/system/lib/libc/musl/arch/emscripten/bits/alltypes.h

Line 95 in 82662c0

#define __LITTLE_ENDIAN 1234

Also, other.test_gen_struct_info still passes on this branch, any ABI break would likely have caused it to fail.

Oh that is rather annoying/strange. It makes things like git grep typedef.*blkcnt_t a little annoying.

Can you add a comment to the end of the emscripen-specific file saying something like this:

// END EMSCRIPTEN-SPECIFIC DEFINITIONS // // Below here are the shared musl definitions. The emscripten-specific definitions above will take precedence // due to the `__DEFINED_` macro system.

Good idea! I added a comment via 4eba06a. We could perhaps consider marking this file with linguist-generated=true in .gitattributes to ensure the diff is hidden by default, though I don't feel strongly about it.

Nah, I don't think this file should change often, and when it does I really want to see the details since this is ABI-defining stuff.

sbc100

Thanks for working on this! LGTM with one final comment.

kleisauke added 2 commits July 23, 2025 14:34

Revise alltypes.h update process. NFC

4ba2500

Align it with musl, i.e. only update `alltypes.h.in` and regenerate `alltypes.h` using a script.

Deduplicate arch/emscripten/bits/alltypes.h{,in} with include/libc/al…

4bb13ba

…ltypes.h.in

This was referenced Jul 23, 2025

Use 64-bit fields for fsblkcnt_t and fsfilcnt_t #24769

Merged

[WIP] Merge musl-1.2.1 into #13006 #13007

Draft

sbc100 reviewed Jul 23, 2025

View reviewed changes

kleisauke added 4 commits July 23, 2025 18:57

Incorporate feedback

4a3bcba

Reinstate __cplusplus conditional

ecefbbd

Merge branch 'main' into revise-musl-alltypes-h

f496d34

Update after af67939

82662c0

sbc100 reviewed Jul 24, 2025

View reviewed changes

kleisauke added 2 commits July 25, 2025 11:36

Merge branch 'main' into revise-musl-alltypes-h

bce77c4

Ensure byte-for-byte identical libc-mt.a

6809780

sbc100 approved these changes Jul 25, 2025

View reviewed changes

Incorporate feedback

4eba06a


		#if defined(__FLT_EVAL_METHOD__) && __FLT_EVAL_METHOD__ == 0

	#if defined(__NEED_wint_t) && !defined(__DEFINED_wint_t)
	typedef __WINT_TYPE__ wint_t;
	#define __DEFINED_wint_t
	#endif

	#if defined(__NEED_wint_t) && !defined(__DEFINED_wint_t)
	typedef unsigned wint_t;
	#define __DEFINED_wint_t
	#endif

	// XXX EMSCRIPTEN: ensure it's always 32-bits even in wasm64
	#if defined(__NEED_blkcnt_t) && !defined(__DEFINED_blkcnt_t)
	typedef int blkcnt_t;
	#define __DEFINED_blkcnt_t
	#endif

	// XXX EMSCRIPTEN: ensure it's always 32-bits even in wasm64
	TYPEDEF int blkcnt_t;
	TYPEDEF int blksize_t;
	TYPEDEF int clock_t;
	TYPEDEF unsigned int dev_t;
	TYPEDEF int suseconds_t;
	TYPEDEF unsigned int wctype_t;

Revise alltypes.h update process. NFC #24773

Are you sure you want to change the base?

Revise alltypes.h update process. NFC #24773

Conversation

kleisauke commented Jul 23, 2025

Uh oh!

sbc100 left a comment

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!

Choose a reason for hiding this comment

Uh oh!

kleisauke Jul 23, 2025 • 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!

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!

sbc100 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Revise `alltypes.h` update process. NFC #24773

Revise `alltypes.h` update process. NFC #24773

kleisauke Jul 23, 2025 •

edited

Loading