Inline documentation patch...

Linus Torvalds torvalds@transmeta.com
Mon Feb 8 21:51:00 GMT 1999


On Mon, 8 Feb 1999, Richard Henderson wrote:
>   * contains computed jump (the point of contention)

Is this the SAME braindamage once more? I thought we had this cleared up
once already and that people were complaining some new reason for gcc
refusing to inline.

The only reason the test for computed jumps was added was due to a gcc bug
(with functions NOT marked inlined), and it's ridiculous and stupid to
then penalize people who _do_ want to inline. 

I (and others) told people what the gcc bug was and how to fix it, but
instead of fixing the bug somebody added MORE buggy code that then
disables functions marked "inline" from being inlined.

Edification for people who didn't see this originally: 

 Basically, the gcc bug was/is that when gcc is used with the flag
"-finline-functions", it will try inline everything that is small enough,
and that shows some _real_ problems when you inline things with static
computed branch pointers.

Instead of fixing "-finline-functions" to just not inline those kinds of
functions, someone decided that the proper solution was to screw up
everybody else instead.

This is getting ridiculous. Just as I was at the stage where I thought
that egcs could be used to compile kernels reliably, somebody apparently
merged some broken gcc-2.8.x code into it.

Back to considering gcc-2.7.2 the only stable compiler out there.

			Linus



More information about the Gcc-patches mailing list