Uh oh!
There was an error while loading.Please reload this page.
- Notifications
You must be signed in to change notification settings - Fork938
Attempt 2 - Fix Missing Git Executable Causing ClusterFuzz Crash#1909
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to ourterms of service andprivacy statement. We’ll occasionally send you account related emails.
Already on GitHub?Sign in to your account
Attempt 2 - Fix Missing Git Executable Causing ClusterFuzz Crash#1909
Uh oh!
There was an error while loading.Please reload this page.
Conversation
This is a second attempt atgitpython-developers#1906 and should resolve:-gitpython-developers#1905-google/oss-fuzz#10600PRgitpython-developers#1906 had the right idea but wrong implementation, and the differences betweenthe ClusterFuzz image that it was supposed to fix and the OSS-Fuzz image wherethe fix was tested led to the issue not being fully resolved.The root cause of the issue is the same: A Git executable is not globallyavailable in the ClusterFuzz container environment where OSS-Fuzz executesfuzz tests.gitpython-developers#1906 attempted to fix the issue by bundling the Git binary and usingGitPython's `git.refresh(<full-path-to-git-executable>)` method to set itinside the `TestOneInput` function of the test harness.However, GitPython attempts to set the binary at import time via its `__init__`hook, and crashes the test if no executable is found during the import.This issue is fixed here by setting the environment variable that GitPythonlooks in before importing it, so it's available for the import. This was testedby setting the `$PATH` to an empty string inside the test files, whichreproduced the crash, then adding the changes introduced here with `$PATH` stillempty, which avoided the crash indicating that the bundled Git executable isworking as expected.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others.Learn more.
Thanks so much for taking charge of this! No worries about taking more steps either :).
797009d
intogitpython-developers:mainUh oh!
There was an error while loading.Please reload this page.
It worked! I'm seeing logs from successful fuzz target runs over the last two hours or so! |
ClusterFuzz runs of the `fuzz_submodule` target have been failingbecause the `git` import was placed before the condition that sets theGit executable path.The order in which `git` is imported matters because it attempts to finda Git executable as the import is loaded (via `refresh()` in`git/__init__.py`.) As pergitpython-developers#1909, we configure the ClusterFuzzenvironment to use a bundled Git executable via the env variablecondition in all fuzz targets.
Uh oh!
There was an error while loading.Please reload this page.
This is a second attempt at#1906 and should resolve:
PR#1906 had the right idea but wrong implementation. The differences between the ClusterFuzz image that it was supposed to fix and the OSS-Fuzz image where it was tested led to the issue not being fully resolved.
The root cause of the issue is the same: A Git executable is not globally available in the ClusterFuzz container environment where OSS-Fuzz executes fuzz tests.
#1906 attempted to fix the issue by bundling the Git binary and using
GitPython's
git.refresh(<full-path-to-git-executable>)
method to set it inside theTestOneInput
function of the test harness.However, GitPython attempts to set the binary at import time via its
__init__
hook, and crashes the test if no executable is found during the import.This issue is fixed here by setting the environment variable that GitPython looks in before importing it, so it's available for the import. This was tested by setting the
$PATH
to an empty string inside the test files, which reproduced the crash, then adding the changes introduced here with$PATH
still empty, which avoided the crash indicating that the bundled Git executable is working as expected.Note: I didn't test this in a full ClusterFuzz environment, so it's worth noting that while it may be possible for this to continue failing because of the difference between environments that the Git executable was produced in vs. where it's run: I don't believe that will be an issue in practice because the guidance for projects (including Git itself) in OSS-Fuzz is to compile in build.sh with the libraries necessary to create the final binary that is executed in ClusterFuzz. The changes here assume that the apt repo version of Git has everything statically built. If, for some reason, it turns out that a static binary isn't what is installed, the alternative will be to build git from source or mock it. But based on my local testing thus far, I think (and hope) that won't be necessary, and this changeset should be sufficient.