New issue
Advanced search Search tips

Issue 884848 link

Starred by 1 user

Issue metadata

Status: Fixed
Owner:
Closed: Nov 20
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Windows
Pri: 2
Type: Bug

Blocking:
issue 884849
issue 883538



Sign in to add a comment

Investigate why the number of PCs varies so wildly in libFuzzer builds on Windows

Project Member Reported by metzman@chromium.org, Sep 17

Issue description

For example, on Linux: base64_decode_fuzzer and base64_encode_fuzzer have about 20k PCs. 
On Windows: decode has 68 and encode has 44k.
When I remove the calls to CHECK and CHECK_EQ from encode, it ends up with about 70 PCs.
I'm guessing linking is not stripping as much dead code as it should (on Linux as well, but that is a separate and known issue with ASAN).

 
Blocking: 883538
Blocking: 884849
Using /O1 instead of /O2 (as done on Linux, thanks to optimize_for_fuzzing) looks promising, in bringing the encode fuzzer to ~29k edges, as opposed to ~25k on Linux. It also reduced the size of all files produces (ie: lib, pdb, and exe)
Status: Fixed (was: Started)
This no longer looks like a problem.

The fuzzers in the example have the following pc counts on the latest build:
base64_decode_fuzzer: 46
base64_encode_fuzzer: 1710

Sign in to add a comment