FatELF: universal binaries for Linux [LWN.net]

FatELF: universal binaries for Linux

Posted Oct 30, 2009 4:31 UTC (Fri) by pj (subscriber, #4506)
Parent article: FatELF: universal binaries for Linux

He totally overlooked the killer app for FatELF: cross-compiling.

One of the major pains of cross-compiling is getting all the library paths sorted - which ones do your tools use vs which ones will what you're building with those tools use, etc. With FatELF, cross-platform development just got a *lot* simpler, because the paths all stay the same, and the tools look in the standard place for the library... and inside the library for the arch they're wanting to link against.

Relatedly, if compilers and linkers are FatELF-aware, they can build fatELFs a bit faster because they only have to do the parsing once, and then object code for all the platforms from the same resultant parse tree.

So the people he should really be appealing to are embedded developers who deal with all this pain _all the time_. The OpenWRT and etc guys will be ecstatic, if only to save them from LD_FLAG hell.

to post comments

FatELF: universal binaries for Linux

Posted Oct 30, 2009 6:14 UTC (Fri) by ringerc (subscriber, #3071) [Link] (1 responses)

"Relatedly, if compilers and linkers are FatELF-aware, they can build fatELFs a bit faster because they only have to do the parsing once, and then object code for all the platforms from the same resultant parse tree."

Unfortunately, that's not true in any language where the use of a preprocessor is standard and commonplace, like C or C++. The preprocessor is a step run before parsing, and the work the preprocessor does may change what the result of parsing is by adding/removing/changing parts of the program text. It's very common in C/C++ to use the preprocessor to handle things like type selection, type size and byte order issues - exactly the sorts of things that'll change depending on target arch.

FatELF: universal binaries for Linux

Posted Nov 4, 2009 16:00 UTC (Wed) by pj (subscriber, #4506) [Link]

Good point. I still think the win as a solution to arch-based-path hell for cross-compiles might be worth it.