Open GoogleCodeExporter opened 9 years ago
are you sure this is not working this way already?
Original comment by konstant...@gmail.com
on 22 May 2014 at 10:30
It does not. If I add llvm-symbolizer to PATH but set
ASAN_OPTIONS=external_symbolizer_path=asdf I get the following warning:
==24065==ERROR: AddressSanitizer: stack-buffer-overflow on address
0x7fffac2c969b at pc 0x4b5748 bp 0x7fffac2c95f0 sp 0x7fffac2c95e8
WRITE of size 1 at 0x7fffac2c969b thread T0
==24065==WARNING: invalid path to external symbolizer!
==24065==WARNING: Failed to use and restart external symbolizer!
#0 0x4b5747 (/usr/local/google/ssd/chrome-git/src/t+0x4b5747)
Original comment by ramosian.glider@gmail.com
on 22 May 2014 at 10:58
Then don't set ASAN_OPTIONS=external_symbolizer_path=asdf !!
Original comment by konstant...@gmail.com
on 22 May 2014 at 11:05
I'm talking about the following scenario.
Most often Chrome developers run their binaries from src/, so we may want to
compile the default external_symbolizer_path into the binary.
However if the working dir is different the devs won't be able to get
symbolized output despite they may have llvm-symbolizer in $PATH. In this case
we may want to check for llvm-symbolizer in $PATH if it isn't in
external_symbolizer_path.
Original comment by ramosian.glider@gmail.com
on 22 May 2014 at 11:11
or we should link chrome with the hermetic symbolizer...
Original comment by konstant...@gmail.com
on 22 May 2014 at 11:15
I disagree with #4. ASAN_OPTIONS=external_symbolizer_path should have a higher
priority than whatever llvm-symbolizer we find in PATH, because ASAN_OPTIONS
are overridable by user. Suppose that the user wants to run his code with a new
version of llvm-symbolizer and runs:
ASAN_OPTIONS=external_symbolizer_path=my/shiny/new/llvm-symolizer ./a.out
(note that missing "b" in llvm-symbolizer). We should fail hard in this case
(hey, the path you've provided is wrong!) rather than silently fallback to
symbolizer in $PATH.
Original comment by samso...@google.com
on 22 May 2014 at 6:51
Original issue reported on code.google.com by
ramosian.glider@gmail.com
on 22 May 2014 at 7:02