Binary size analysis

[Tao Liu]

Hi

    Today I use "src/tools/binary_size/run_binary_size_analysis.py  --library src/out/Release/content_shell --destdir /tmp/report" to profile content_shell size, but  I got this :

#tools/binary_size/run_binary_size_analysis.py --library out/Release/content_shell --destdir /tmp/report

addr2line: /usr/bin/arm-linux-gnueabi-addr2line

nm: /usr/bin/arm-linux-gnueabi-nm

38.5%: Looked up 94800 symbols (4054 collisions, 0 disambiguations where 0.0% succeeded) - 226.9 lookups/s.

100.0%: Looked up 246500 symbols (14077 collisions, 0 disambiguations where 0.0% succeeded) - 4968.5 lookups/s.

232462 symbols in the results.

WARNING:root:There are sections with 155894 nodes. Results might be unusable.

Writing 34764536 bytes json

Report saved to /tmp/report/index.html

My chromium is m49.0.2603 and binary format is Arm v7, /tmp/report is too big to upload. Any ideas?

Thanks,

Tao, Liu

[Daniel Bratell]

So addr2line probably failed for many symbols and filed them in "No source".  It looks like it managed to handle 100k of the 250k symbols but I can't tell why it failed the rest. In the past I've had to build my own addr2line to support newer DWARF formats but I think it failed completely then.

Have you looked at the output? Maybe you can find a pattern in the symbols that ended up in "No Source" and maybe it's still usable to you.

See if you can get anything useful when running addr2line by hand on the binary. You should have an nm file in the output directory with the symbol addresses to use.

(Improvements are welcome!)

/Daniel

--

/* Opera Software, Linköping, Sweden: CET (UTC+1) */