On Tue, Oct 15, 2024 at 12:44:53PM +0200, Patrick Steinhardt wrote: [snip] > > I have read some patches yesterday, I feel quite strange that we need to > > make repetition. Could we provide a header file which requires the users > > who need to use the reftable library to implement the interfaces? > > > > reftable_strbuf_addf(void *buf, char *fmt, va_list ap); > > > > Thus, we could reuse "strbuf_addf" to implement this interface in Git. > > As for libgit2, could we let it implement these interfaces? Although I > > have never read the source code of libgit2, I think there should be some > > code which could be reuse to implement these interfaces? > > > > However, I do not know the context. Maybe the above is totally wrong. If > > so, please ignore. > > The thing is that we'll have repetition regardless of what we end up > doing: > > - We could either have repetition once in the reftable library, > reimplementing `struct strbuf`. This can then be reused by every > single user of the reftable library. > > - Or we can have repetition for every single user of the reftable > library. For now that'd only be Git and libgit2, but we'd still have > repetition. > > The second kind of repetition is way worse though, because now every > user of the reftable library has a different implementation of a type > that is as basic as a buffer. These _must_ behave the exact same across > implementations or we will hit issues. So I'd rather have the repetition > a single time in the reftable library such that all users of the library > will behave the same rather than having downstream users copy the > implementation of `struct strbuf` and making it work for their library. > Yes. I agree with you it is worse to let every downstream to implement the interfaces. I know the motivation here, we want to make the whole reftable library be independent of the Git which allows the downstream to easily use the reftable library. > Also, due to the nature of `struct strbuf` not allowing for allocation > failures we'd already have diverging behaviour. In Git you would never > hit error code paths for allocation failures, whereas every library user > potentially can. > > So we really have to treat the reftable code base special. If we want to > be a good citizen and be a proper upstream for projects like libgit2 we > don't really have much of a choice than to detangle it from libgit.a. If > we don't we may be saving 20 lines of code, but we make everybody elses > life harder. > Yes. And I do not think this is a problem right now. Thanks for this wonderful explanation. > Patrick Jialuo
