premature exits bork cache

[John Keeping]

On Fri, Feb 26, 2016 at 01:29:21PM +0100, Jason A. Donenfeld wrote:
> When caching is turned on, if we exit(0) early from a cmd, the cache
> won't be properly finalized (or even sent), and we get a server error
> because of blank output. I just fixed this bug with redirect:
> 
> https://git.zx2c4.com/cgit/commit/?id=e9cbdf64632fbe64d6b0c4974ac947ef954938e7
> 
> This is fine, because it's called only from 1 function (three times),
> and that function then returns, so execution doesn't continue after
> the redirect. It simply gives control back to the caching layer that
> does things properly now.
> 
> But I noticed we have this issue with HTTP HEAD method:
> 
> https://git.zx2c4.com/cgit/tree/ui-shared.c#n695
> 
> Here we're calling exit(0). The motivation, I imagine, was that we
> want to retain the same handlers, but just return the headers for the
> HEAD method. Unfortunately, this borks the cache layer for the same
> reasons as described above. I'm not yet sure the cleanest way to fix
> this. Suggestions?

We set cfg.nocache in cgit.c for HEAD requests so I think we're already
okay in this case, although it is a subtle interaction.