On Mon, Mar 30, 2026 at 12:04:37PM +0100, Stuart Henderson wrote:
> On 2026/03/30 12:25, Theo Buehler wrote:
> > On Mon, Mar 30, 2026 at 11:09:50AM +0100, Stuart Henderson wrote:
> > > Some ports are happy to _build_ in parallel, but fail during "make
> > > fake". For example, vim and php.
> > > 
> > > We used to have a separate PARALLEL_INSTALL variable controlling
> > > whether parallel flags were passed through there, added in r1.920,
> > > removed in r1.1510.
> > > 
> > > I'm not convinced that parallel fake is enough of a win to be worth
> > > restricting the number of ports that can use parallel build so I'd like
> > > to remove it. (Apart from rare semi-broken ports which do a chunk of
> > > compilation during "make fake", it's mostly going to be filesystem
> > > access which doesn't parallelize brilliantly on OpenBSD at present
> > > anyway).
> > 
> > One of the ports where this will really hurt is x1//qt5/qtbase which
> > rebuilds itself in its entirety during fake.
> 
> Oof.
> 
> > I don't know how many rust ports with DPB_PROPERTIES=parallel there are
> > but if any of them sets MODCARGO_INSTALL_TARGET_PATHS, it will likely
> > build a non-trivial amount of stuff as well.
> 
> Perhaps it does want an easier way to select on/off than adding to
> FAKE_FLAGS then. (I don't know what the default should be).
> 
> re rust: lang/deno, mail/stalwart, www/chromium, www/iridium set
> DPB_PROPERTIES. (ungoogled-chromium doesn't currently have rust parts).
> stalwart is the only one setting MODCARGO_INSTALL_TARGET_PATHS.
> However, in general rust ports do seem like they might be a good
> candidate for parallel, and MODCARGO_INSTALL_TARGET_PATHS is fairly
> common.
> 
> > So while I'm ok with this in principle, I think this needs to run
> > through a bulk or two to see if it doesn't add a significant amount of
> > build time.
> 
> Makes sense, I think I'll try to find some time later during lock to
> do some testing.
> 
> 

You could always disable by default and add it by hand in the corresponding
ports.  After all, both PARALLEL_MAKE_FLAGS and ALL_FAKE_FLAGS are user-visible
variables, with nothing preventing people from handling them manually.

If there's just a small number of ports, it's simpler than re-adding another
flag.