Ühel kenal päeval, E, 13.12.2021 kell 10:19, kirjutas Marek Szuba:
> On 2021-12-09 15:04, Michał Górny wrote:
> 
> > Why do you need to use random name in the first place?  We have
> > full
> > control over T, so why not just hardcode a good name?
> 
> Having discussed the matter with eclass maintainers on IRC, they are
> not 
> entirely sure whether using a static name in this context is entirely
> safe. There were also concerns about making this change too
> aggressive 
> given it affects all supported EAPIs. Therefore, we have decided to
> play 
> it safe and stick as closely to old behaviour as possible, at least
> for now.
> 
> Anyway, merged a moment ago.

Actually I kind of preferred a static name over straight mktemp,
because emktemp supported other cases than a pure mktemp usage does.
But I don't know if it could ever clash things in some weird
situations. I think they won't, but I don't know if PMS guarantees that
or just happens how portage works right now (e.g. the postrm currently
happening in a separate ._unmerge directory path for $T; multilib
postinst happening sequentially, etc).

Thinking it through again a bit, straight mktemp can't be worse than a
static name anyways (provided mktemp exists, which emktemp handled..),
so we're good there, but provided you or someone thinks through the
corner-cases, I'm in favor of a static name if it doesn't have any
trouble.


Mart