platform/architecture
Dear Taco, (moved from the context mailing list) On Thu, Jan 27, 2011 at 07:25, Taco Hoekwater wrote:
On 01/27/11 00:21, Mojca Miklavec wrote:
However ... this discussion was already being held long ago ... there is no reason why luatex could not be as smart as "bash --version" is and aware of the architecture it was compiled for (that is not necessary the platform it is running on; an obvious case being i386 luatex running on 64-bit Mas OS X for example). Somebody would have to write some simple code to check for that though.
I'm reviving this discussion since my next question would be a request for Hans to add support for mips to ConTeXt. But that is a highly tricky business. I don't know any better how to do it than config.guess does (please close your eyes :) mips:Linux:*:* | mips64:Linux:*:*) eval $set_cc_for_build sed 's/^ //' << EOF >$dummy.c #undef CPU #undef ${UNAME_MACHINE} #undef ${UNAME_MACHINE}el #if defined(__MIPSEL__) || defined(__MIPSEL) || defined(_MIPSEL) || defined(MIPSEL) CPU=${UNAME_MACHINE}el #else #if defined(__MIPSEB__) || defined(__MIPSEB) || defined(_MIPSEB) || defined(MIPSEB) CPU=${UNAME_MACHINE} #else CPU= #endif #endif EOF It actually runs the compiler and tests for some flags. You could do the same, except that this script is well maintained and has (hopefully) proper support for many exotic platforms. I see that config.guess resides in the mpost & luatex repository already. The latest version can probably be fetched from http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.guess... but that is not really important in most cases. To me it seems that the script does a really decent job (except when it doesn't :) Honestly, I'm not sure what exactly it does in case of cross-compilers; I would need your help with that. There are two other problems that I'm aware of, but I'm almost able to fix both and can try to send patches upstream to see if they accept them. If they don't, it can still be fixed in luatex source repository. 1.) Configuration for Apple is slightly wrong in case of cross-compilers, but I can either submit a patch to the project or you could patch the file in your repository in case that they reject the patch (I also need somebody's assistance for writing the patch, but at least I know exactly what has to be checked). 2.) In case of Debian (including mips) it doesn't check if the system is actually 64-bit or not. In my opinion this should be fixed. Apart from that, the script returns three (four) parameters, separated with a dash: CPU_TYPE-MANUFACTURER-OPERATING_SYSTEM 1.) cpu type (uname -m; UNAME_MACHINE) 2.) manufacturer 3.) os (uname -s; UNAME_SYSTEM) 4.) version; optional; UNAME_RELEASE (and no dash separating system & version) The versions are almost arbitrary or absent. For linux it is "-gnu" or "-$(libc)" (I wonder why the dash in this case). The call luatex --version could return exactly the same string as config.guess or as bash --version (with slight modifications where needed). But it would have to know at least about the first and third variable, possibly about version as well (when present and when it makes sense). The function os.uname returns a table with sysname, machine, release, version, and nodename I would suggest to provide one similar set of values, providing information about the build architecture as obtained by config.guess. Honestly, I don't mind if luatex uses "darwin" instead of "macosx". Actually, I probably prefer darwin. If ConTeXt wants to further modify these names, it should be free to do so, but LuaTeX should be as close to original as possible. The number 11 (config.guess returns x86_64-apple-darwin11) is useful to have as the information about backward compatibility, at least in case of mac. But I'm not yet sure how to check for that during compilation. I can try to figure it out. What do you think? Mojca
Hi Mojca, I am not sure I understand this post, not even after lots of rereading. I know that it is about os.platform/os.type based on the earlier discussion, but that is about where it ends for me, sorry. On 08/18/11 01:34, Mojca Miklavec wrote:
Dear Taco,
(moved from the context mailing list)
On Thu, Jan 27, 2011 at 07:25, Taco Hoekwater wrote:
On 01/27/11 00:21, Mojca Miklavec wrote:
However ... this discussion was already being held long ago ... there is no reason why luatex could not be as smart as "bash --version" is and aware of the architecture it was compiled for (that is not necessary the platform it is running on; an obvious case being i386 luatex running on 64-bit Mas OS X for example). Somebody would have to write some simple code to check for that though.
I'm reviving this discussion since my next question would be a request for Hans to add support for mips to ConTeXt. But that is a highly tricky business. I don't know any better how to do it than config.guess does (please close your eyes :)
mips:Linux:*:* | mips64:Linux:*:*) eval $set_cc_for_build sed 's/^ //' << EOF >$dummy.c #undef CPU #undef ${UNAME_MACHINE} #undef ${UNAME_MACHINE}el #if defined(__MIPSEL__) || defined(__MIPSEL) || defined(_MIPSEL) || defined(MIPSEL) CPU=${UNAME_MACHINE}el #else #if defined(__MIPSEB__) || defined(__MIPSEB) || defined(_MIPSEB) || defined(MIPSEB) CPU=${UNAME_MACHINE} #else CPU= #endif #endif EOF
It actually runs the compiler and tests for some flags. You could do the same, except that this script is well maintained and has (hopefully) proper support for many exotic platforms.
I see that config.guess resides in the mpost & luatex repository already. The latest version can probably be fetched from http://git.savannah.gnu.org/gitweb/?p=config.git;a=blob_plain;f=config.guess... but that is not really important in most cases.
To me it seems that the script does a really decent job (except when it doesn't :) Honestly, I'm not sure what exactly it does in case of cross-compilers; I would need your help with that.
There are two other problems that I'm aware of, but I'm almost able to fix both and can try to send patches upstream to see if they accept them. If they don't, it can still be fixed in luatex source repository.
1.) Configuration for Apple is slightly wrong in case of cross-compilers, but I can either submit a patch to the project or you could patch the file in your repository in case that they reject the patch (I also need somebody's assistance for writing the patch, but at least I know exactly what has to be checked).
2.) In case of Debian (including mips) it doesn't check if the system is actually 64-bit or not. In my opinion this should be fixed.
Apart from that, the script returns three (four) parameters, separated with a dash: CPU_TYPE-MANUFACTURER-OPERATING_SYSTEM 1.) cpu type (uname -m; UNAME_MACHINE) 2.) manufacturer 3.) os (uname -s; UNAME_SYSTEM) 4.) version; optional; UNAME_RELEASE (and no dash separating system & version)
The versions are almost arbitrary or absent. For linux it is "-gnu" or "-$(libc)" (I wonder why the dash in this case).
The call luatex --version could return exactly the same string as config.guess or as bash --version (with slight modifications where needed). But it would have to know at least about the first and third variable, possibly about version as well (when present and when it makes sense).
The function os.uname returns a table with sysname, machine, release, version, and nodename I would suggest to provide one similar set of values, providing information about the build architecture as obtained by config.guess.
Honestly, I don't mind if luatex uses "darwin" instead of "macosx". Actually, I probably prefer darwin. If ConTeXt wants to further modify these names, it should be free to do so, but LuaTeX should be as close to original as possible.
The number 11 (config.guess returns x86_64-apple-darwin11) is useful to have as the information about backward compatibility, at least in case of mac. But I'm not yet sure how to check for that during compilation. I can try to figure it out.
What do you think?
Mojca _______________________________________________ dev-luatex mailing list dev-luatex@ntg.nl http://www.ntg.nl/mailman/listinfo/dev-luatex
On Thu, Aug 18, 2011 at 15:48, Taco Hoekwater wrote:
Hi Mojca,
I am not sure I understand this post, not even after lots of rereading. I know that it is about os.platform/os.type based on the earlier discussion, but that is about where it ends for me, sorry.
So I suggest to assign a 15-30-minutes slot somewhere during the meeting (it doesn't have to be on the official schedule), so that I can explain my request in detail, explain my idea how to solve it and to leave time for some brainstorming (with other hackers that will be present and interested). It would be nice to solve this properly before luatex version 1 comes out. Thanks, Mojca
participants (2)
-
Mojca Miklavec -
Taco Hoekwater