There are various issues with how wrap malloc is being used, from the nonsense of the variable names to the lack of flags when building C programs. It also happens that android builds without jemalloc fail (and even if they wouldn't fail at build time, they'd fail at run time on devices without enough free space on the internal flash).

This makes things clearer (IMHO), and fixes Android build without jemalloc.

Without the nspr part, and with a temporary workaround until the nspr part lands

Comment on attachment 555371 [details] [diff] [review] Cleanup wrap malloc and its use on Android. NSPR part [Checked in: Comment 8] Review of attachment 555371 [details] [diff] [review]: ----------------------------------------------------------------- ::: nsprpub/configure.in @@ +3180,5 @@ > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=malloc,--wrap=calloc,--wrap=valloc,--wrap=free,--wrap=realloc,--wrap=memalign" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=__builtin_new,--wrap=__builtin_vec_new,--wrap=__builtin_delete,--wrap=__builtin_vec_delete" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=PR_Free,--wrap=PR_Malloc,--wrap=PR_Calloc,--wrap=PR_Realloc" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=strdup,--wrap=strndup" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=posix_memalign,--wrap=malloc_usable_size" Can you maybe do these as a heredoc or something? WRAP_LDFLAGS=`cat<<EOF ${WRAP_LDFLAGS} ... ... ... EOF` ?

Comment on attachment 555368 [details] [diff] [review] Cleanup wrap malloc and its use on Android Review of attachment 555368 [details] [diff] [review]: ----------------------------------------------------------------- ::: build/unix/elfhack/Makefile.in @@ +74,5 @@ > $(NULL) > > libs:: $(CSRCS:.c=.$(OBJ_SUFFIX)) > > +WRAP_FLAGS= You probably meant WRAP_LDFLAGS= here, right? ::: configure.in @@ +7473,5 @@ > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=malloc,--wrap=calloc,--wrap=valloc,--wrap=free,--wrap=realloc,--wrap=memalign" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=__builtin_new,--wrap=__builtin_vec_new,--wrap=__builtin_delete,--wrap=__builtin_vec_delete" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=PR_Free,--wrap=PR_Malloc,--wrap=PR_Calloc,--wrap=PR_Realloc" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=strdup,--wrap=strndup" > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=posix_memalign,--wrap=malloc_usable_size" Same comment here as in the NSPR patch. ::: security/coreconf/OS2.mk @@ +70,5 @@ > HIGHMEM_LDFLAG = -Zhigh-mem > endif > > ifndef NO_SHARED_LIB > +WRAP_LDFLAGS = This is an NSS file...

(In reply to Ted Mielczarek [:ted, :luser] from comment #5) > > +WRAP_FLAGS= > > You probably meant WRAP_LDFLAGS= here, right? Yes > ::: security/coreconf/OS2.mk > > +WRAP_LDFLAGS = > > This is an NSS file... Oops (In reply to Ted Mielczarek [:ted, :luser] from comment #4) > Comment on attachment 555371 [details] [diff] [review] > Cleanup wrap malloc and its use on Android. NSPR part > > Review of attachment 555371 [details] [diff] [review]: > ----------------------------------------------------------------- > > ::: nsprpub/configure.in > @@ +3180,5 @@ > > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=malloc,--wrap=calloc,--wrap=valloc,--wrap=free,--wrap=realloc,--wrap=memalign" > > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=__builtin_new,--wrap=__builtin_vec_new,--wrap=__builtin_delete,--wrap=__builtin_vec_delete" > > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=PR_Free,--wrap=PR_Malloc,--wrap=PR_Calloc,--wrap=PR_Realloc" > > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=strdup,--wrap=strndup" > > + WRAP_LDFLAGS="${WRAP_LDFLAGS} -Wl,--wrap=posix_memalign,--wrap=malloc_usable_size" > > Can you maybe do these as a heredoc or something? > WRAP_LDFLAGS=`cat<<EOF > ${WRAP_LDFLAGS} ... > ... > ... > EOF` > ? Putting it on several lines doesn't work because the variable then contains newlines, and that messes up the sed used to generate autoconf.mk. WRAP_LDFLAGS=`cat <<EOF | xargs echo ${WRAP_LDFLAGS} ... ... EOF` would work, but it's somehow ugly, isn't it? Plus the EOF needs not to be indented. Except if we do WRAP_LDFLAGS=`cat <<" EOF" | xargs echo ... EOF

Without the typo and the nss file.

Checking in configure; /cvsroot/mozilla/nsprpub/configure,v <-- configure new revision: 1.313; previous revision: 1.312 done Checking in configure.in; /cvsroot/mozilla/nsprpub/configure.in,v <-- configure.in new revision: 1.315; previous revision: 1.314 done Checking in config/autoconf.mk.in; /cvsroot/mozilla/nsprpub/config/autoconf.mk.in,v <-- autoconf.mk.in new revision: 1.45; previous revision: 1.44 done Checking in config/rules.mk; /cvsroot/mozilla/nsprpub/config/rules.mk,v <-- rules.mk new revision: 3.84; previous revision: 3.83 done Checking in lib/tests/Makefile.in; /cvsroot/mozilla/nsprpub/lib/tests/Makefile.in,v <-- Makefile.in new revision: 1.25; previous revision: 1.24 done Checking in pr/tests/Makefile.in; /cvsroot/mozilla/nsprpub/pr/tests/Makefile.in,v <-- Makefile.in new revision: 1.68; previous revision: 1.67 done

And without the nspr workaround

http://hg.mozilla.org/projects/build-system/rev/14497b5c651e http://hg.mozilla.org/projects/build-system/rev/cf4d258fd0f6