This is the mail archive of the glibc-linux@ricardo.ecn.wfu.edu mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

2.2.1 setjmp/tst-setjmp.c: Possible regression from glibc-2.2.0


This is just a heads up: I haven't analyzed this much yet, because I'm
too tired to do so.

In glibc-2.2, and all previous versions I know of, building and testing
a glibc configured with --enable-omitfp worked perfectly. Despite the
claims in INSTALL:

,----
|                                                                    The
|      extra optimization doesn't gain you much, it may provoke compiler
|      bugs
`----

my distinctly unscientific measurements show that the gains, at least on
IA32, are sufficient (a little under ten percent, perhaps because of the
lower insn cache hit?) to be worth it, and -fomit-frame-pointer in GCC
is pretty damned robust; my entire system is built with it, for
starters, and except for Mozilla everything is happy with it. (And
that's just because Mozilla's NSPR threading library appears to expect a
frame pointer.)


But in 2.2.1, setjmp/tst-setjmp.c, when built with -fomit-frame-pointer,
dumps core at program termination time, indicating that something has
smashed the stack somewhere. Yet there've been no immediately obvious
changes in 2.2 -> 2.2.1 that could've caused this, at least not to csu
or setjmp. (Maybe the change is somewhere else.)

I haven't tracked it down yet, but before I do that and report this as a
glibc bug, has anyone seen this before, maybe with an earlier version of
glibc? (I'd feel a right twit reporting this as a new glibc bug if it's
actually old.)

I'll try to track this down tomorrow. Debugging with no frame pointer,
oh joy...

-- 
`Anyhow, that pipe dream doesn't say anything about the question you
 asked.  (I am planning for a career in politics.)' --- Mark Mitchell
                                                      on the GCC list

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]