638056 - #error "The cacheFlush support is missing on this platform." on mips, s390 and ia64

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Mike Hommey [:glandium]]

Previously, one didn't need to --disable something to successfully build on exotic architectures. This particular failure seems to be related to

ifeq (,$(filter-out powerpc sparc,$(TARGET_CPU)))
here -> http://mxr.mozilla.org/mozilla-central/source/js/src/Makefile.in#378

It looks like the logic is reversed here, and that it should use a filter instead. It also looks like this should be enclosed in a ifdef ENABLE_something.

Comment 1

[Mike Hommey [:glandium]]

Attachment: Avoid "The cacheFlush support is missing on this platform" error on exotic platforms

I tested this on s390, and this allows to build a working engine. I will test other architectures soon.

Comment 2

[Landry Breuil (:gaston)]

(In reply to comment #1)
> Created attachment 520157 [details] [diff] [review]
> Avoid "The cacheFlush support is missing on this platform" error on exotic
> platforms
> 
> I tested this on s390, and this allows to build a working engine. I will test
> other architectures soon.

Yes please, tested working fine on OpenBSD/amd64 (which correctly enters in assembler/) and OpenBSD/sparc64 (which doesnt gets there). OpenBSD/mips64el test on the way.

Previously, i was using 
+ifneq (,$(filter-out sparc64,$(TARGET_CPU)))
 CXXFLAGS += -DUSE_SYSTEM_MALLOC=1 -DENABLE_ASSEMBLER=1 -DENABLE_JIT=1
+else
+CXXFLAGS += -DUSE_SYSTEM_MALLOC=1
+endif

using --disable-tracejit --disable-methodjit on sparc64 was not enough.

but mike's patch looks ways saner. Maybe enclose it within ENABLE_JIT test ? anyone can test on i386/amd64 with --disable-tracejit ?

Btw, forcing -DENABLE_ASSEMBLER=1 -DENABLE_JIT=1 at the end of Makefile.in regardless of the arch looks weird.

And i think https://bugzilla.mozilla.org/show_bug.cgi?id=620368 & https://bugzilla.mozilla.org/show_bug.cgi?id=630854 could be marked as dupes of this bug.

Comment 3

[Mike Hommey [:glandium]]

(In reply to comment #2)
> Btw, forcing -DENABLE_ASSEMBLER=1 -DENABLE_JIT=1 at the end of Makefile.in
> regardless of the arch looks weird.

Yet, it apparently builds and works fine, provided the js engine actually works (cf. jsvalue problems on ia64 and sparc64).

Comment 6

[Landry Breuil (:gaston)]

Confirmed producing a working js binary on OpenBSD/mips64el. Well, at least it runs like on sparc64, independently of the jsvalue problems we're likely to meet on mips64 too.

Comment 7

[Mike Hommey [:glandium]]

(In reply to comment #2)
> using --disable-tracejit --disable-methodjit on sparc64 was not enough.
> 
> but mike's patch looks ways saner. Maybe enclose it within ENABLE_JIT test ?
> anyone can test on i386/amd64 with --disable-tracejit ?

I think --disable-something bugs should be separated.

Comment 8

[Chris Leary [:cdleary] (not checking bugmail)]

Comment on attachment 520157 [details] [diff] [review]
Avoid "The cacheFlush support is missing on this platform" error on exotic platforms

Don't actually know the syntax, but the semantics look good! Feel free to r? someone who knows the syntax better if you're at all concerned about it.

Comment 9

[Mike Hommey [:glandium]]

http://hg.mozilla.org/mozilla-central/rev/59771590e920

Comment 11

[Anton Shterenlikht]

With the patch firefox-4.0.1 builds for me
on FreeBSD ia64 9.0 -current.

However, it dumps core on startup.