I use the following script to cross compile x86-64 -> x86 on Linux: http://hg.mozilla.org/users/bsmedberg_mozilla.com/configure-scripts/file/1e492a582e34/configure-m32 This has broken sometime recently: jsautocfg.h has host values for everything, instead of target values, in particular: #define JS_BYTES_PER_WORD 8L This manifests itself as a build error: ../../../src/js/src/jsatom.h:69: error: size of array ‘arg’ is negative Which is on the line: JS_STATIC_ASSERT(sizeof(JSAtom *) == JS_BYTES_PER_WORD);
Is there an understood ideal way to fix this? If so, can someone describe what needs to happen?
I'm not sure this should block -- why? Jim, could you take a look here?
It's a fairly recent regression, and makes it difficult for me personally to do static-checking of JIT builds (since we don't have JIT on x86-64). But I think it's fairly well in hand now that the configured uintN types have landed.
Does this checkin http://hg.mozilla.org/mozilla-central/rev/525e42406396 interfere with this bug? I've tested to build a x86_64 build on Mac OS X after this checkin and now the contents of jsautocfg.h looks different than before.
Fixed by bug 269538. I'm now able to build js/src standalone by doing: CC="gcc -arch x86_64" CXX="g++ -arch x86_64" AR=ar ../mozilla-central/js/src/configure --target=x86_64-apple-darwin (not sure when that requirement for AR crept in.)
Hm, Bug bug 269538 is no reopened and the patch is backed out due to Windows TUnit bustage...
Fixed on trunk and branch by bug 269538