This is not the right way to load default values. There is a config var cache in git_repository. Can you please use that to store the default value, and then load it into the diff context from the cache? Thanks!

@vmg In defense of this code, it is copying code that I wrote (immediately above it in the file) that does the same thing for a bool value. Diff looks at core.symlinks, core.ignorestat, core.filemode, core.trustctime, and with this PR diff.context. Also, implicitly it uses core.ignorecase. Should we move all of those into the repository cvar cache? Should pure boolean config values have explicit entries in git_cvar_value or can we just use generic -1, 0, and 1 without expanding that enum?

By the way, there are a ton of other config options that we could probably be checking also (and growing all the time). This diff.context is just one of probably 10+ other settings that we could (should?) be applying. Of course, if we do that, we need to start being more thorough in the test suite to isolate ourselves from the global and system config settings!

@arrbee: If you feel that all these options are not used enough to deserve caching, then this helper that "looks up config var and returns it or a default without setting an error message" needs to be part of the config code. Looks like a very common operation to perform, so let's implement it where it belongs.

@vmg - sounds good. I think the answer is somewhere in the middle. core.symlinks core.filemode and core.ignorecase should almost certainly be moved into the cvar cache. Maybe all of the core... ones, actually. diff.context and some of the other diff... configs are fairly uncommon, so a config helper that took a default and did not set an error on a failed lookup would be a useful thing.

@linquize Do you want to take this on? If so, feel free! Otherwise, I'm happy to grab your commit as a start and do the rest of the things that @vmg is suggesting... Just let me know.

Yes, you can start your work based on this commit.

Merged as part of #1499. I had to rebase, so this commit ended up getting merged as a5df71c instead, but it should still be credited to you. Thanks!