> slightly size negative (which surprised me).

That would surprise me too, so I patched in your change to take a look. Looks like the sizes are:

$ tar tvzf clang-349417-1.tgz | grep bin/clang
-rwxr-xr-x chrome-bot/chrome-bot 104989048 2018-12-18 04:03 bin/clang
lrwxrwxrwx chrome-bot/chrome-bot         0 2018-12-18 04:03 bin/clang++ -> clang
lrwxrwxrwx chrome-bot/chrome-bot         0 2018-12-18 04:03 bin/clang-cl -> clang

$ tar tvzf clang-349417-2.tgz | grep bin/clang
lrwxrwxrwx pcc/primarygroup        0 2018-12-21 12:21 bin/clang-cl -> clang
lrwxrwxrwx pcc/primarygroup        0 2018-12-21 12:21 bin/clang++ -> clang
-rwxr-xr-x pcc/primarygroup 98156600 2018-12-21 12:21 bin/clang

Are you sure that you were looking at the right binaries?

Yes, reasonably sure.

Note how the "before" dir has the .so files an the "after" file has the -8 symlink, which suggests that the former is downloaded by the package, and the latter is built locally.

To get the numbers, I looked at the downloaded clang (I was at 347874-5, before yesterday's roll). Doing that again, that's 104,312,552 bytes, and 104312552 >> 20 is 99, which matches the "before" number in comment 0. It also roughly matches your "before" number.

I then ran package.py with patch set 1 from https://chromium-review.googlesource.com/c/chromium/src/+/1387395 and got the "After:" numbers. I do not know why I'd get different numbers than you for this step, but the clang-8 symlink and the absence of .so files suggests to me that I didn't hold it completely wrong.

Weird.

Ah, the copy of clang-8 in my third_party/llvm-build directory is approximately the same size as yours:

$ ls -lh third_party/llvm-build/Release+Asserts/bin/clang-8
-rwxr-xr-x 1 pcc primarygroup 105M Dec 21 12:17 third_party/llvm-build/Release+Asserts/bin/clang-8

But if I strip it I get the same size as in my tarball:

$ strip -o foo third_party/llvm-build/Release+Asserts/bin/clang-8
$ ls -l foo
-rwxr-xr-x 1 pcc primarygroup 98156600 Dec 21 13:01 foo

And we are indeed stripping files before adding them to the tarball:
https://cs.chromium.org/chromium/src/tools/clang/scripts/package.py?l=410
So that explains it.

Hm, I did add a sys.exit(1) to package.py in https://chromium-review.googlesource.com/c/chromium/src/+/1387395/1/tools/clang/scripts/package.py but it was after the strip step. Not sure why the strip step apparently didn't run over here then.

(But looking at this I see that with CLANG_PLUGIN_SUPPORT=OFF we can stop passing -x to strip on mac; I'll update my CL.)

Mac:

http://commondatastorage.googleapis.com/chromium-browser-clang/index.html?path=Mac/

clang-349417-1.tgz 44.33MB
clang-349417-2.tgz 41.81MB

2.5MB / 5.6% smaller compressed size.

Unpacked:
bin/clang 93.6MB -> 88.5MB
libBlinkGCPlugin.dylib was 868kB, libFindBadConstructs.dylib was 302kB

6.27MB / 6.7% uncompressed size.



Linux:

http://commondatastorage.googleapis.com/chromium-browser-clang/index.html?path=Linux_x64/
clang-349417-1.tgz 75.97MB
clang-349417-2.tgz 73.66MB

2.31 MB / 3% smaller compressed size.

Unpacked:
bin/clang 105MB -> 99.1MB

libBlinkGCPlugin.so was 1.7MB, libFindBadConstructs.so was 567kB

3.67MB / 3.5% uncompressed size.




(why is the linux package so much bigger? because it's much more data: 258.6MB unpacked, vs 141.1MB, 117.5MB more. Why? We bundle lld on Linux but not mac, that's 71.4MB, this explains all but 46.1MB. llvm-14 are 18.4MB, leaving 27.7MB. The lib/ folder is 18.8MB larger (runtimes for android), and the rest is probably slightly larger binaries, maybe due to the GOT/PLT indirection. But the linux binary shrinking less kind of disproves that gues.)

Err, linux should've read:

(105 - 99.1 + 1.7 + 0.567) / 105

8.167MB / 7.7%


So uncompressed shrank similarly, and compressed the percentage is smaller because clang is a smaller percentage of the whole package.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de

commit 25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de
Author: Nico Weber <thakis@chromium.org>
Date: Thu Jan 03 06:09:34 2019

Build clang without plugin support.

See bug for details.

Bug:  917404 , 256342 
Change-Id: Ic5fd959f23e77b9194af4faecceeefffe0b924ae
Reviewed-on: https://chromium-review.googlesource.com/c/1387395
Commit-Queue: Nico Weber <thakis@chromium.org>
Reviewed-by: Reid Kleckner <rnk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#619595}
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/build/config/clang/BUILD.gn
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/third_party/blink/renderer/BUILD.gn
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/tools/clang/blink_gc_plugin/CMakeLists.txt
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/tools/clang/blink_gc_plugin/tests/test.py
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/tools/clang/plugins/CMakeLists.txt
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/tools/clang/plugins/tests/test.py
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/tools/clang/pylib/clang/plugin_testing.py
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/tools/clang/scripts/package.py
[modify] https://crrev.com/25a9c5ec90f0bc3df2fdc827d5b36b83bc8520de/tools/clang/scripts/update.py