Skip to contentSkip to conversation

diffutils: musl has working malloc() and realloc()

PATCH In review Andr? Draszik 3 participants 6 messages Jan 8, 2020 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

PATCH v2 2/2 diffutils: musl has working malloc() and realloc()

diffutils tries to run a test program to determine if malloc() and realloc() work, i.e. whether they fail if given 0 for size. running a test program doesn't work when cross-compiling, so it defaults to assuming non- working malloc() / realloc()

Given they work correctly in musl, we can specify the test-result, and avoid having to use diffutil's replacement functions.

---
v2:
* switch to CACHED_CONFIGUREVARS instead of EXTRA_OECONF

---
 recipes-extended/diffutils/diffutils_2.8.1.bb | 5 +++++
 1 file changed, 5 insertions(+)
recipes-extended/diffutils/diffutils_2.8.1.bb +5 -0
index 4aad28f..680911b 100644
@@ -16,6 +16,11 @@ SRC_URI = "${GNU_MIRROR}/diffutils/diffutils-${PV}.tar.gz \
 SRC_URI[md5sum] = "71f9c5ae19b60608f6c7f162da86a428"
 SRC_URI[sha256sum] = "c5001748b069224dd98bf1bb9ee877321c7de8b332c8aad5af3e2a7372d23f5a"
 
+CACHED_CONFIGUREVARS_libc-musl = "\
+    jm_cv_func_working_malloc=yes \
+    jm_cv_func_working_realloc=yes \
+"
+
 do_configure_prepend () {
 	chmod u+w ${S}/po/Makefile.in.in
 }
Signature · 1 line
2.23.0.rc1
#2 lore raw

recipes-extended/diffutils/diffutils_2.8.1.bb SRC_URI = "${GNU_MIRROR}/diffutils/diffutils-${PV}.tar.gz \ · line 22

Quoted 4 lines · Andr? Draszik
recipes-extended/diffutils/diffutils_2.8.1.bb +4 -0
+CACHED_CONFIGUREVARS_libc-musl = "\
+    jm_cv_func_working_malloc=yes \
+    jm_cv_func_working_realloc=yes \
+"

Why is this musl-specific? Does the configure explicitly detect glibc and know it's good, or should we do this for glibc too?

Ross

#3 lore raw
Quoted 8 lines · Ross Burton

recipes-extended/diffutils/diffutils_2.8.1.bb SRC_URI = "${GNU_MIRROR}/diffutils/diffutils-${PV}.tar.gz \ · line 22

Quoted 4 lines · Andr? Draszik
recipes-extended/diffutils/diffutils_2.8.1.bb +4 -0
+CACHED_CONFIGUREVARS_libc-musl = "\
+    jm_cv_func_working_malloc=yes \
+    jm_cv_func_working_realloc=yes \
+"

Why is this musl-specific? Does the configure explicitly detect glibc and know it's good, or should we do this for glibc too?

That's musl specific, because it's a property of the C-library in use.

E.g. somebody could be using this with an outdated vendor(?) toolchain using an old glibc which could behave incorrectly...

Cheers,
Andre'

#4 lore raw
Quoted 7 lines · Andr? Draszik
Quoted 6 lines · Ross Burton
Quoted 4 lines · Andr? Draszik
recipes-extended/diffutils/diffutils_2.8.1.bb +4 -0
+CACHED_CONFIGUREVARS_libc-musl = "\
+    jm_cv_func_working_malloc=yes \
+    jm_cv_func_working_realloc=yes \
+"
Quoted 8 lines
Quoted 2 lines

Why is this musl-specific? Does the configure explicitly detect glibc and know it's good, or should we do this for glibc too?

That's musl specific, because it's a property of the C-library in use.

E.g. somebody could be using this with an outdated vendor(?) toolchain using an old glibc which could behave incorrectly...
...

The jm_cv_func_working_malloc test is:

    main ()
    {
      exit (malloc (0) ? 0 : 1);
    }

The C standard says that this is implementation defined, both results are permitted.

glibc has the same behaviour forever,
musl switched to the glibc behaviour a few years ago.

Quoted 2 lines

Cheers,
Andre'

cu
Adrian

#5 lore raw
Quoted 23 lines · Adrian Bunk
Quoted 7 lines · Andr? Draszik
Quoted 6 lines · Ross Burton
Quoted 4 lines · Andr? Draszik
recipes-extended/diffutils/diffutils_2.8.1.bb +4 -0
+CACHED_CONFIGUREVARS_libc-musl = "\
+    jm_cv_func_working_malloc=yes \
+    jm_cv_func_working_realloc=yes \
+"
Quoted 8 lines
Quoted 2 lines

Why is this musl-specific? Does the configure explicitly detect glibc and know it's good, or should we do this for glibc too?

That's musl specific, because it's a property of the C-library in use.

E.g. somebody could be using this with an outdated vendor(?) toolchain using an old glibc which could behave incorrectly...
...

The jm_cv_func_working_malloc test is:

    main ()
    {
      exit (malloc (0) ? 0 : 1);
    }
Quoted 5 lines

The C standard says that this is implementation defined, both results are permitted.

glibc has the same behaviour forever,
musl switched to the glibc behaviour a few years ago.

Thanks for the reminder. Final note, Back in the days, uClibc had a compile time option to select one behaviour or the other.

Cheers,
A.

#6 lore raw
In reply to Ross Burton
Quoted 8 lines · Ross Burton

recipes-extended/diffutils/diffutils_2.8.1.bb SRC_URI = "${GNU_MIRROR}/diffutils/diffutils-${PV}.tar.gz \ · line 22

Quoted 4 lines · Andr? Draszik
recipes-extended/diffutils/diffutils_2.8.1.bb +4 -0
+CACHED_CONFIGUREVARS_libc-musl = "\
+    jm_cv_func_working_malloc=yes \
+    jm_cv_func_working_realloc=yes \
+"

Why is this musl-specific? Does the configure explicitly detect glibc and know it's good, or should we do this for glibc too?

Sorry, I was too quick replying here... It's simply defaulting to no because of cross-compilation, no matter what libc is in use.

I wanted to be conservative, but Adrian reminded me of actual issue here...

Cheers,
Andre'