Subject: Re: remove default of STRIPFLAG=-s ?
To: None <tech-toolchain@NetBSD.org>
From: David Laight <firstname.lastname@example.org>
Date: 02/11/2004 22:02:32
On Wed, Feb 11, 2004 at 04:27:58PM +0300, Valeriy E. Ushakov wrote:
> On Thu, Feb 12, 2004 at 00:00:42 +1100, Luke Mewburn wrote:
> > I've been considering for a while that we should remove the
> > default build setting of STRIPFLAG=-s (which causes programs to
> > be stripped (with strip(1)) as they're installed (by install(8))
> BTW, is there an opensource analog to SysV mcs(1) (manage comment
> section)? It can be stripped by adding -R .comment to STRIPFLAG, but
> mcs(1) has a nice -c flag:
> -c Compress the contents of the comment section of the
> ELF object files. All duplicate entries are removed.
> The ordering of the remaining entries is not dis-
> The savings, though, are, arguably, minuscule - e.g. libc.so has a
> .comment section of about 30K.
But only because we don't have #ident lines in all the header files.
Do that and 'mcs -c' is your friend.
IMHO we should build without removing any symbols, then strip for
installation. The assumption then is that the build tree has the
symbols in it - maybe the unstripped binaries should be available
for the release CDs?
I have had to rebuild programs because they get stripped during build.
That is bad news. Also removing local symbols can give apparrantly
incorrect stack back-traces, so isn't particularly useful.
It it also possible to do a reasonable amount of debugging with just
the standard symbol table - after all the most usable feature of gdb is
the 'bt' command on a core dump.
David Laight: email@example.com