Skip to contentSkip to conversation

coreutils: fix build with 64bit time_t on 32bit arches

PATCH André Draszik 1 participant 2 messages Jan 20, 2020 – Jan 27, 2020 2 messages 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

ping

Quoted 57 lines · Andr? Draszik

From: André Draszik <[email protected]>

Quoted 7 lines

../../coreutils-6.9/lib/mktime.c: In function 'ydhms_diff':
../../coreutils-6.9/lib/mktime.c:105:52: error: size of array 'a' is negative
105 | #define verify(name, assertion) struct name { char a[(assertion) ? 1 : -1]; }
| ^
../../coreutils-6.9/lib/mktime.c:168:3: note: in expansion of macro 'verify'

  168 |   verify (long_int_year_and_yday_are_wide_enough,
      |   ^~~~~~

coreutils tries to determine of mktime() from the c-library works fine, but when cross-compiling, it unconditionally assumes buggy behaviour, in which case it will use its own replacement implementation. These basically are tests for (g)libc implementation bugs that have existed in ancient times but aren't relevant anymore.

The alternative implementation makes various assumptions which don't hold true (and doesn't compile anymore). While more recent versions of coreutils have fixed those assumptions, it's quite hard to update the code here without potential licensing issues.

Given mktime() works fine in musl and glibc these days, we can work-around the issues by simply relying on the c-libraries mktime() implementation, avoiding all problems with coreutils' alternative implementation.

I've veryfied this by running the tests both on my musl target, and my glibc (v2.29) host.

This also reduces code-size slightly.

The ac_cv_func_working_mktime variable here corresponds to the gl_cv_func_working_mktime variable in more recent versions of coreutils (gnulib), which is being set in oe-core in meta/site.

---
 recipes-core/coreutils/coreutils_6.9.bb | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
recipes-core/coreutils/coreutils_6.9.bb +3 -1
index 080100d..f6017ec 100644
@@ -33,7 +33,9 @@ SRC_URI[md5sum] = "c9607d8495f16e98906e7ed2d9751a06"
 SRC_URI[sha256sum] = "89c2895ad157de50e53298b22d91db116ee4e1dd3fdf4019260254e2e31497b0"
 
 EXTRA_OECONF += "ac_cv_func_getgroups_works=yes \
-                 ac_cv_func_strcoll_works=yes"
+                 ac_cv_func_strcoll_works=yes \
+                 ac_cv_func_working_mktime=yes \
+                 "
 
 # acl is not a default feature
 #

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

#3 lore raw

ping.

meta-gplv2 is broken against OE/master without this...

A.

Quoted 62 lines · André Draszik

ping

Quoted 57 lines · Andr? Draszik

From: André Draszik <[email protected]>

Quoted 7 lines

../../coreutils-6.9/lib/mktime.c: In function 'ydhms_diff':
../../coreutils-6.9/lib/mktime.c:105:52: error: size of array 'a' is negative
105 | #define verify(name, assertion) struct name { char a[(assertion) ? 1 : -1]; }
| ^
../../coreutils-6.9/lib/mktime.c:168:3: note: in expansion of macro 'verify'

  168 |   verify (long_int_year_and_yday_are_wide_enough,
      |   ^~~~~~

coreutils tries to determine of mktime() from the c-library works fine, but when cross-compiling, it unconditionally assumes buggy behaviour, in which case it will use its own replacement implementation. These basically are tests for (g)libc implementation bugs that have existed in ancient times but aren't relevant anymore.

The alternative implementation makes various assumptions which don't hold true (and doesn't compile anymore). While more recent versions of coreutils have fixed those assumptions, it's quite hard to update the code here without potential licensing issues.

Given mktime() works fine in musl and glibc these days, we can work-around the issues by simply relying on the c-libraries mktime() implementation, avoiding all problems with coreutils' alternative implementation.

I've veryfied this by running the tests both on my musl target, and my glibc (v2.29) host.

This also reduces code-size slightly.

The ac_cv_func_working_mktime variable here corresponds to the gl_cv_func_working_mktime variable in more recent versions of coreutils (gnulib), which is being set in oe-core in meta/site.

---
 recipes-core/coreutils/coreutils_6.9.bb | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
recipes-core/coreutils/coreutils_6.9.bb +3 -1
index 080100d..f6017ec 100644
@@ -33,7 +33,9 @@ SRC_URI[md5sum] = "c9607d8495f16e98906e7ed2d9751a06"
 SRC_URI[sha256sum] = "89c2895ad157de50e53298b22d91db116ee4e1dd3fdf4019260254e2e31497b0"
 
 EXTRA_OECONF += "ac_cv_func_getgroups_works=yes \
-                 ac_cv_func_strcoll_works=yes"
+                 ac_cv_func_strcoll_works=yes \
+                 ac_cv_func_working_mktime=yes \
+                 "
 
 # acl is not a default feature
 #