296138 - Use lxLite to compress binaries

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Peter Weilbacher]

With the hints from Paul Ratcliffe in the newsgroup I found the correct switches
for lxLite to better strip and compress the binaries but not garble the
functionality of chkdll32:

   lxlite /ydd /yxd /d

I get even higher compression ratios using

   lxlite /mr3 /ml1 /mf2 /ydd /yxd /d

but then it never finishes for gklayout.dll and actually the result in the ZIP
file is minimal while I guess there might even be a negative effect on slow
machines to uncompress the files again.

Comment 1

[Peter Weilbacher]

Attachment: reactivate lxLite in strip.cmd

This activates the call to lxlite in strip.cmd, we win about one MB... Hmm,
would this mean lxLite need to get a build requirement for Mozilla? It seems
that if I remove lxLite from the path make -C xpinstall/packager runs through
anyway without error.

Comment 2

[Felix Miata]

Do this and we can't run ft2lib 1.0 without undoing lxlite on applicable files
any more. For proper development QA we need to be able to run Moz/FF with and
without ft2lib, and this is too complicated for under ft2lib 2+, probably much
like having to remove lxlite. I have no need for anything in v2 that isn't in
v1. Maybe lxlite on releases, but not nightlies please.

Comment 3

[Peter Weilbacher]

For QA you should certainly not rely on ft2lib 1.0 any more! It had a lot of
bugs that are fixed in the latest version. And with ft2lib 2.0 it's as easy as
flipping a switch in the (yes, urgs) registry. I can do some research and come
up with a script to do that more easily than with regedit/2 if you want.

Comment 4

[Peter Weilbacher]

Attachment: Speed and size statistics

I just had some fun measuring the effects of various lxLite parameters. The table in the attached file basically shows that the parameters in the patch (attachment 184967 [details] [diff] [review]) give the best compromize in terms of speed and (download) size while keeping chkdll32 working. Well, its not such a big effect for startup speed compared to no lxLite but it buys half a MB of download size.

The size columns give the package size with zip -9rD and unpacked dir size, the startup speed was measured as mean+-stddev and median using the mozilla/tools/performace/startup scripts on a laptop which I once throttled to 13% of its CPU speed (10 samples each) and once used without throttling (24 samples).

Comment 5

[Mike Kaply [:mkaply]]

Comment on attachment 184967 [details] [diff] [review]
reactivate lxLite in strip.cmd

r=mkaply - as long as chkdll32 still works, I'm ok with this.

Comment 6

[Mike Kaply [:mkaply]]

fix checked in.

Comment 7

[Peter Weilbacher]

Comment on attachment 184967 [details] [diff] [review]
reactivate lxLite in strip.cmd

Small OS/2 packaging change only, long trunk tested.

Comment 8

[Darin Fisher]

Comment on attachment 184967 [details] [diff] [review]
reactivate lxLite in strip.cmd

a=darin on behalf of drivers