This is the mail archive of the mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [ANNOUNCEMENT] Updated: libtool-devel-20030216, libltdl-20030216

Ralf Habacker wrote:
o speed problem with the internal "win32_libid()" routine is partially
fixed -- approx 35% speedup.  Ralf Habacker is working on some
modifications to the official fileutils distribution (specifically,
file.exe) that should allow an additional 8% speedup without additional
changes to libtool.

Charles, thanks for your efforts with this libtool stuff and I have no problem
to gone with this already started work, but let me ask one question: Is the
needed effort hacking the fileutils distro justifiabled for only 8% speedup,
where you have already got 35% speedup. I'm worry about, that this speedup will
not be noticeable.

Well, remember that my testing methodology is very rough. I just ran win32_libid() on the entire contents of my /lib and /bin dirs. That isn't a normal usage pattern -- I'm *only* checking win32_libid() and not any of the other activities that are important in a linkphase. So it is unclear (a) *exactly* how much faster a real link would be given this change, and (b) how much faster a super-`file` version would be, in a real production environment.

It's possible that *in actual use*, your proposed file changes would account for more than 8% -- and my changes would account for less than 35%.

But, those were the numbers I had.

Should we not instead leave this as it is and concentrate on the way replacing
import libraries with links as much as can be, which will give much more benefit
because this case is already as fast as it could be handled by your script (x86
DLL) ?

It's up to you -- that's why I coded the libtool changes so that your improvements could be a "drop in" fix. I think it'd be "nice" is file could tell me that a given file was an import lib or a (regular) lib, but it isn't necessary.

There's certainly no rush.

As I stated before, I have already a libtool hack which covers mostly of the
needed stuff (a little work is necessary: integrating in recent libtool and fix
uninstall target)

I look forward to seeing that...


-- Unsubscribe info: Bug reporting: Documentation: FAQ:

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