Firefox 20.0.1 build problems

[Gabriele Bulfon]

Hi,

I'm working to have latest Firefox (20.0.1) packaged in our upcoming distro of illumos,

Sonicle XStream Desktop.

I know there is no support for Solaris like OS, but I'm just checking to see if you can help me.

I could apply a couple of quick temporary patches to move on, like these:

jsmath.cpp

--------------

had to take out a piece of (old I think) code about Solaris, MOZ_DOUBLE_IS_NEGZERO is not

defined anywhere:

--- mozilla-release/js/src/jsmath.cpp Wed May 8 12:48:54 2013

+++ mozilla-release/js/src/jsmath.cpp Wed May 8 12:50:34 2013

@@ -195,14 +195,6 @@

}

#endif

-#if defined(SOLARIS) && defined(__GNUC__)

- if (x == 0) {

- if (MOZ_DOUBLE_IS_NEGZERO(y))

- return js_copysign(M_PI, x);

- if (y == 0)

- return x;

- }

-#endif

return atan2(x, y);

}

Maybe there is some more correct solution.

vm/Stack.cpp

----------------

This piece of code complains about converting "base_" class into "char *" on mincore(..), on Solaris.

I could take out all of the ifdef section and fall back to the last section:

--- mozilla-release/js/src/vm/Stack.cpp Wed May 8 13:13:08 2013

+++ mozilla-release/js/src/vm/Stack.cpp Wed May 8 13:13:47 2013

@@ -782,57 +782,11 @@

size_t

StackSpace::sizeOf()

{

-#if defined(XP_UNIX)

/*

- * Measure how many of our pages are resident in RAM using mincore, and

- * return that as our size. This is slow, but hopefully nobody expects

- * this method to be fast.

- *

- * Note that using mincore means that we don't count pages of the stack

- * which are swapped out to disk. We really should, but what we have here

- * is better than counting the whole stack!

- */

-

- const int pageSize = getpagesize();

- size_t numBytes = (trustedEnd_ - base_) * sizeof(Value);

- size_t numPages = (numBytes + pageSize - 1) / pageSize;

-

- // On Linux, mincore's third argument has type unsigned char*.

-#ifdef __linux__

- typedef unsigned char MincoreArgType;

-#else

- typedef char MincoreArgType;

-#endif

-

- MincoreArgType *vec = (MincoreArgType *) js_malloc(numPages);

- int result = mincore(base_, numBytes, vec);

- if (result) {

- js_free(vec);

- /*

- * If mincore fails us, return the vsize (like we do below if we're not

- * on Windows or Unix).

- */

- return (trustedEnd_ - base_) * sizeof(Value);

- }

-

- size_t residentBytes = 0;

- for (size_t i = 0; i < numPages; i++) {

- /* vec[i] has its least-significant bit set iff page i is in RAM. */

- if (vec[i] & 0x1)

- residentBytes += pageSize;

- }

- js_free(vec);

- return residentBytes;

-

-#elif defined(XP_WIN)

- return (commitEnd_ - base_) * sizeof(Value);

-#else

- /*

* Return the stack's virtual size, which is at least an upper bound on its

* resident size.

*/

return (trustedEnd_ - base_) * sizeof(Value);

-#endif

}

#ifdef DEBUG

build_config

=========

I just tried adding my own lines for Solaris:

--- mozilla-release/ipc/chromium/src/build/build_config.h Wed May 8 13:45:40 2013

+++ mozilla-release/ipc/chromium/src/build/build_config.h Wed May 8 13:47:38 2013

@@ -19,6 +19,8 @@

#define OS_MACOSX 1

#elif defined(__linux__) || defined(ANDROID)

#define OS_LINUX 1

+#elif defined(__sun)

+#define OS_SOLARIS 1

#elif defined(__DragonFly__)

#define OS_DRAGONFLY 1

#elif defined(__FreeBSD__)

@@ -42,7 +44,7 @@

// For access to standard POSIX features, use OS_POSIX instead of a more

// specific macro.

-#if defined(OS_MACOSX) || defined(OS_LINUX) || defined(OS_BSD)

+#if defined(OS_MACOSX) || defined(OS_LINUX) || defined(OS_BSD) || defined(OS_SOLARIS)

#define OS_POSIX 1

#endif

This way, build goes on for a long time, then breaking on the main problem I'm asking you:

/sources/userlands/xstream-userland-gate/components/firefox/build/i86/_virtualenv/bin/python /sources/userlands/xstream-userland-gate/components/firefox/mozilla-release/config/pythonpath.py \

-I/sources/userlands/xstream-userland-gate/components/firefox/mozilla-release/other-licenses/ply \

/sources/userlands/xstream-userland-gate/components/firefox/build/i86/dist/sdk/bin/header.py -I/sources/userlands/xstream-userland-gate/components/firefox/mozilla-release/netwerk/dns -I../../dist/idl /sources/userlands/xstream-userland-gate/components/firefox/mozilla-release/netwerk/dns/nsPIDNSService.idl -d .deps/nsPIDNSService.h.pp -o _xpidlgen/nsPIDNSService.h

/sources/userlands/xstream-userland-gate/components/firefox/mozilla-release/config/rules.mk:1581: *** unterminated variable reference. Stop.

make[5]: Leaving directory `/sources/userlands/xstream-userland-gate/components/firefox/build/i86/netwerk/dns'

Looking at the source of rules.mk, looks like there is a kind of syntax error occurring while "dns"

Makefile includes it. I could not find any reason for this....
Any idea????

Thanks for any help!
Gabriele.

[Atsushi Shimmura]

Hi, I'm building Firefox 20.0.1 on Solaris 11.1,and I met this case so
much. Any case , I clean up build tree and try "gmake -f client.mk
build",for some reason ,it goes out. Please show me your mozconfig.

my mozconfig:
mk_add_options MOZ_OBJDIR=@TOPSRCDIR@/../objdir-tb3~
ac_add_options --prefix=$HOME
ac_add_options --enable-application=browser
ac_add_options --enable-optimize
ac_add_options --disable-necko-wifi
ac_add_options --disable-tests
ac_add_options --disable-ipc
ac_add_options --disable-crashreporter
ac_add_options --enable-official-branding
ac_add_options --disable-dbm
ac_add_options --enable-default-toolkit=cairo-gtk2

> Makefile includes it. I could not find any reason for this....
> Any idea????
> Thanks for any help!
> Gabriele.

--
Atsushi