Re: Proposal/Discussion: Turning parts of Git into libraries

[Jeff King <peff@peff.net>]

On Fri, Feb 17, 2023 at 02:49:51PM -0800, Emily Shaffer wrote:

> > Personally, I'd like to see some sort of standard error type (whether
> > integral or not) that would let us do more bubbling up of errors and
> > less die().  I don't know if that's in the cards, but I thought I'd
> > suggest it in case other folks are interested.
> 
> Yes!!! We have talked about this a lot internally - but this is one
> thing that will be difficult to introduce into Git without making
> parts of the codebase a little uglier. Since obviously C doesn't have
> an intrinsic to do this, we'll have to roll our own, which means that
> manipulating it consistently at function exits might end up pretty
> ugly. So hearing that there's interest outside of my team to come up
> with such a type makes me optimistic that we can figure out a
> neat-enough solution.

Here are some past discussions on what I thought would be a good
approach to error handling. The basic idea is to replace the "pass a
strbuf that people shove error messages into" approach with an error
context struct that has a callback. And that callback can then stuff
them into a strbuf, or report them directly, or even die.

This thread sketches out the idea, though sadly I no longer have the
more fleshed-out patches I mentioned there:

  https://lore.kernel.org/git/20160927191955.mympqgylrxhkp24n@sigill.intra.peff.net/

And then the sub-thread starting here discusses a similar approach:

  https://lore.kernel.org/git/20171103191309.sth4zjokgcupvk2e@sigill.intra.peff.net/

It does mean passing a "struct error_context" just about everywhere.
Though since the context doesn't change very much and most calls are
just forwarding it along, it would probably also be reasonable to have a
thread-local global context, and push/pop from it (sort of a poor man's
dynamic scoping).

One thing that strategy doesn't help with, though, that your
libification might want: it's not very machine-readable. The error
reporters would still fundamentally be working with strings. So a
libified process can know "OK, writing this ref failed, and I have some
error messages in a buffer". But the calling code can't know specifics
like "it failed because we tried to open file 'foo' and it got EPERM".
We _could_ design an error context that stores individual errno values
or codes in a list, but having each caller report those specific errors
is a much bigger job (and ongoing maintenance burden as we catalogue and
give an identifier to each error).

-Peff