Re: nasm + libtool + shared lib

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


|
| | Still no clue what might cause the problems on linux.
|
| I think I found the cause of the problem
| (although I still don't understand why this can
| result in a segfault...).
|
| Somwhere in the nasm manual, in a section about
| building shared libraries, I found a remark
| that ebx is used to store the address of the
| global offset table.
| In a code snippet, I saw that ebx was saved before
| any further computations were done.
| So I simply push ebx in the function prologue, and
| pop it in the epilogue. Works just fine.
|
| I don't understand, however, why the compiler
| can assume that the function does not modify ebx.
| Any ELF/shared library experts?

This [1] explains the behaviour (and also a strange
thing with ebx that happened to me in a piece
of inline assembler once: gcc did not preserve
ebx himself, although it was listed in the
cloblist, and then segfaulted, see [2]).

Makes me wonder how compiler and os independent
our assembler code really is.
What if another
os/compiler uses a different base register
that must not be changed?
But then, it's assembler code, so what did
we expect :)

| I will try to replace all inline assembler with
| external nasm files in the next weeks.

Merged into main.

[1]
http://linux4u.jinr.ru/usoft/WWW/www_debian.org/Documentation/elf/node21.html
[2] http://lists.debian.org/debian-gcc/2002/07/msg00094.html
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBFHde/rP0cdKF/ToRAriwAJ9Qo8cPMlAavEroJqa0OYA4c+wT7QCg0n5n
cvdLthSd0vK7lfWlmW1Ct7g=
=Jnb2
-----END PGP SIGNATURE-----