Re: [RFC PATCH 0/2] gitweb: die_error (error handling) improvements

From: Jakub Narebski <jnareb@gmail.com>
Date: Dec 13, 2010, 07:55 UTC
Message-ID: <201012130855.23013.jnareb@gmail.com>
In-Reply-To: <4D058228.7040905@eaglescrag.net>

On Mon, 13 Dec 2010, J.H. wrote:

> On 12/12/2010 04:46 PM, Jakub Narebski wrote:

Show 10 quoted lines

> > The following two patch series changes improve error / exception
> > handling in gitweb, preparing the way for gitweb output caching, but
> > useful even without it.
> > 
> > I'm sending this patch series early to gather feedback on possible
> > ways of improving error / exception handling in gitweb.
> 
> Personally, instead of another band-aid over this problem, and adding
> (or further legitimizing) goto statements inside gitweb I'd much *MUCH*
> rather we actually put in the work to actually clean this up.

That's not band-aid, that's using Perl exception mechanism.  Gitweb
uses die_error() like one would ordinarily use "die".

> This is the direction I'm heading in, which I mentioned in an earlier
> e-mail.

Well, then how do you want to handle errors? Note that die_error calls are sometimes nested quite deep in the call stack, so using return value to denote errors and checking it is rather out of question: it would significantly increase complexity of code for no gain.

Nevertheless I'll take a look how it is solved in other web applications
written in Perl, like SVN::Web or CPAN Hubble.

Show 5 quoted lines

> There are a *LOT* of disadvantages to the eval mechanism in perl.  It's
> the standard but gitweb is getting more and more complex, and eval is
> simplistic.  Couple that with the complexity and uncertainty that things
> like goto add to the code, I would *MUCH* rather not see this series go
> in, as I think it is the wrong approach to fixing this.

eval / die is not like goto, but like exception mechanism in other languages. I'd prefer to use Try::Tiny or TryCatch, but we have this "no extra dependencies" policy for gitweb.

-- 
Jakub Narebski
Poland