[aur-dev][PATCH] config: allow reading both the proto file and the modified config
nodivbyzero at gmail.com
Tue Apr 17 03:16:56 UTC 2018
I think you're right.
I looked one more time and found guard there:
Thanks for clarification.
On Mon, Apr 16, 2018 at 1:41 AM, Eli Schwartz <eschwartz at archlinux.org> wrote:
> On 04/12/2018 02:22 PM, Nodiv Byzero wrote:
>> The point is that there are few places where config_load() gets called
>> in the sequential order (the same request) and every time it loads and
>> parses data.
> Are these also places where if (!isset($AUR_CONFIG)) fails to actually
> fulfill its sole reason to exist?
> Not that it matters... if for some reason we are loading the config very
> fast and scribbling over each other, because $AUR_CONFIG is not yet set
> when we begin the second config_load() execution, checking if $config is
> already loaded won't help -- it's a local variable.
> So, question: can you show me code where $config would *not* be `false`?
> Because I think this proposition would be either a no-op or a reason to
> file a severely major upstream php bug complaining that the variables
> are leaking all over the floor.
> That being said, yes, there is a race condition where you call
> config_load() twice, and they both set $AUR_CONFIG.
> Elevating $config and $default_config to globals, *then* using them as
> additional file access caches to micro-optimize one or two disk ops,
> seems wasteful to me. It's not like this is being run in a tight loop.
> If any effort should be spent to fix this, rather than introducing
> painful, non-obvious code, we should introduce some sort of proper cache.
> Eli Schwartz
> Bug Wrangler and Trusted User
More information about the aur-dev