OS X 10.6 (Snow Leopard) build issues

[ddp]

Hi,

New user here. I'm trying to build Jazz from git on 10.6.2 with
Gambit 4.5.3 and Xcode 3.2.1 (gcc 4.2.1). The first issue is that
Gambit builds itself 32-bit by default though it defaults to producing
64-bit binaries. So without doing anything, you blow up in your first
link step trying to link 64-bit binaries against a 32-bit
libgambit.a. I rebuilt Gambit 64-bit mode to solve that:

bash$ CFLAGS="-arch x86_64" LDFLAGS="-arch x86_64" ./configure --
enable-single-host --prefix=/usr/local/gambit

Alternatively you can add an "-m32" in the linker-options definition
in kernel/runtime/build.scm if you need to use 32-bit Gambit.

But wait, there's more... So now I'm stuck here:

% make
make kernel
; creating ./bin...
; creating ./bin/build...
; creating ./bin/build/products...
; creating ./bin/build/products/kernel...
; creating ./bin/build/kernel...
; creating ./bin/build/kernel/syntax...
; creating ./bin/build/kernel/runtime...
; generating ./bin/.configuration...
; generating ./bin/build/kernel/_architecture.scm...
; compiling ./bin/build/kernel/_architecture...
; compiling ./kernel/syntax/header...
; compiling ./kernel/syntax/macros...
; compiling ./kernel/syntax/expansion...
; compiling ./kernel/syntax/features...
; compiling ./kernel/syntax/declares...
; compiling ./kernel/syntax/primitives...
; compiling ./kernel/syntax/syntax...
; compiling ./kernel/syntax/runtime...
; compiling ./kernel/runtime/base...
; compiling ./kernel/runtime/common...
; compiling ./kernel/runtime/build...
; compiling ./kernel/runtime/settings...
; compiling ./kernel/runtime/install...
; compiling ./kernel/runtime/digest...
; compiling ./kernel/runtime/unit...
; compiling ./kernel/runtime/setup...
; generating ./bin/build/products/kernel/_product.scm...
; generating ./bin/build/products/kernel/_main.scm...
; compiling ./bin/build/products/kernel/_product...
; compiling ./bin/build/products/kernel/_main...
; creating link file...
; linking executable...
make core
; compiling core/base/_base...
; compiling core/base/syntax/_syntax...
; compiling core/base/syntax/macros...
; compiling core/base/runtime/_runtime...
; compiling core/base/runtime/boolean...
; compiling core/base/runtime/keyword...
; compiling core/base/runtime/list...
; compiling core/base/runtime/exception...
; compiling core/base/runtime/error...
; compiling core/base/runtime/reader...
; compiling core/base/runtime/serial...
; compiling core/base/runtime/string...
; compiling core/base/runtime/symbol...
; compiling core/base/runtime/syntax...
; compiling core/base/runtime/unspecified...
; compiling core/base/runtime/vector...
; compiling core/class/_class...
; compiling core/class/syntax/_syntax...
; compiling core/class/syntax/error...
; compiling core/class/syntax/object...
; compiling core/class/syntax/class...
; compiling core/class/syntax/define-class...
; compiling core/class/syntax/define-method-expander...
; compiling core/class/syntax/define-method...
; compiling core/class/syntax/classes...
; compiling core/class/runtime/_runtime...
; compiling core/class/runtime/runtime...
; compiling core/class/runtime/output-hook...
; compiling core/class/runtime/output...
; compiling core/exception/_exception...
; compiling core/exception/syntax/classes...
; compiling core/exception/runtime/exception...
; compiling core/exception/runtime/error...
; compiling core/generic/_generic...
; compiling core/generic/syntax/_syntax...
; compiling core/generic/syntax/expander...
; compiling core/generic/syntax/generic...
; compiling core/generic/syntax/specific...
; compiling core/generic/runtime/_runtime...
; compiling core/generic/runtime/specific...
; compiling core/generic/runtime/generic...
; compiling core/module/_module...
; compiling core/module/syntax/_syntax...
; compiling core/module/syntax/classes...
; compiling core/module/syntax/dialect-syntax...
; compiling core/module/syntax/dialect...
; compiling core/module/syntax/walker...

It's been compiling walker for a good 45 minutes now. I'm skeptical
that it's ever going to finish. Is this perhaps a known gcc bug?
Apple also ships gcc 4.0.1 (again, this is 4.2.1). Any opinion on
that or any other workaround would be appreciated...

Thanks!

Derrell

[ddp]

A follow-up to my last post.

I tried falling back to gcc-4.0 and that seems to fix it though
gcc-4.0 does not default to x86_64 and thus I had to rebuild Gambit 32-
bit. However with that done, it's well past walker now. So I
conclude this is a bug in gcc-4.2.1.

Derrell

[Guillaume Cartier]

Indeed. GCC 4.2.x has severe problems handling large files.

A couple points :

First, we reduced walker.scm's size from 240K down to 185K. This should help a bit in all cases but compile times will still be pretty high.

Second, a test was added in the Jazz kernel to automatically disable SINGLE_HOST when using GCC version 4.2.x. This should solve the problem and make it so you can still build Gambit using --enable-single-host and only Jazz's files will have it disabled. Simples tests showed that for Jedi, not have SINGLE_HOST only had about a 5% performance impact so barely noticeable.

Finally, Jazz's build system supports turning on or off Gambit's debugging support in compiled code as they can have a big impact on compilation times.

By default it uses those flags to fully support debugging : debug-environments?: #t debug-location?: #t

By default they should stay turned on so new users have full debugging support. But for seasoned users, I suggest having 2 build configurations : one with those 2 debug flags set to #f and used to run Jedi and one with debugging info and also possibly optimize?: #f for running your Jazz applications with full debugging support.

Personally I have a setup with 2 Jazz git repositories :

- jazz : built without any debugging info and only containing stable Jazz code from the public repository

- jazz_dev : containing all my Jazz changes with full debugging info and optimize?: #f

With this setup, I usually run Jedi from the jazz repo and use it to remotely debug projects using the jazz_dev kernels. More on that setup later as I plan on making that setup into a standard profile included with jazz.

Timings results on Snow Leopard to compile walker.scm:

with SINGLE_HOST : 1h30

with SINGLE_HOST disabled : 3mn

with SINGLE_HOST disabled and no debugging info: 10s !!! so 490x faster

A full rebuild of jedi with SINGLE_HOST disabled and no debugging info is about 11mn.

Guillaume

Derrell

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Jazz Scheme" group.
To post to this group, send email to jazzs...@googlegroups.com
To unsubscribe from this group, send email to jazzscheme+...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/jazzscheme?hl=en
-~----------~----~----~----~------~----~------~--~---