From: Stuart Henderson <stu@spacehopper.org>
Subject: Re: Poison ccache and sccache
To: "Anthony J. Bentley" <bentley@openbsd.org>
Cc: Klemens Nanni <kn@openbsd.org>, ports@openbsd.org
Date: Thu, 10 Sep 2026 14:54:41 +0100

On 2026/09/10 04:48, Anthony J. Bentley wrote:
> Klemens Nanni writes:
> > I'm not convinced we should help the port(er) use ccache(1) directly.
> >
> > Rather all usage should go through USE_CCACHE, which does warn if the
> > package is missing.
> 
> USE_CCACHE is a user setting, so ports aren't allowed to toggle it.
> The only option within a port is to add ccache to BUILD_DEPENDS or to
> disable ccache in the upstream code.

yep.

> Of course disabling is better, but sometimes there is no knob, so we
> patch ccache away. I can imagine a situation where it's too hard to
> patch away (I'm thinking of the handful of ports that still use doxygen
> despite our best efforts). Avoiding a hidden dep is the most important
> thing since it hurts bulks.

I've not seen a ccache configure check that would be at all difficult
to remove. Based on how it works (compiler wrapper/override) it always
ought to be simpler than something with logic that decides whether or
not to process files with some program.

> Here's a better version that encourages disabling ccache first.
> 
> --- infrastructure/mk/bsd.port.mk
> +++ infrastructure/mk/bsd.port.mk
> @@ -452,7 +452,8 @@ CCACHE_DIR ?= ${WRKOBJDIR_${PKGPATH}}/.ccache
>  MAKE_ENV += CCACHE_DIR=${CCACHE_DIR} ${CCACHE_ENV}
>  CONFIGURE_ENV += CCACHE_DIR=${CCACHE_DIR}
>  # scons cleans the environment when calling the compiler
> -COMPILER_WRAPPER += env CCACHE_DIR=${CCACHE_DIR} ${CCACHE_ENV} ccache
> +COMPILER_WRAPPER += env CCACHE_DIR=${CCACHE_DIR} ${CCACHE_ENV} \
> +	${LOCALBASE}/bin/ccache
>  .  if !exists(${LOCALBASE}/bin/ccache)
>  ERRORS += "Fatal: USE_CCACHE is set, but ccache is not installed."
>  .  endif
> @@ -2783,6 +2784,12 @@ ${_WRKDIR_COOKIE}:
>  	@${_PBUILD} mkdir -p ${WRKDIR} ${WRKDIR}/bin
>  	@${_wrap_install_commands}
>  # poison some common binaries unless the relevant BUILD_DEPENDS is used
> +.if empty(_BUILD_DEP:Mdevel/ccache)
> +	@printf '#!/bin/sh\n\
> +		echo "*** $$0 was called; disable it or add an explicit dependency ***" >&2\n\

I think I'd probably say "was called directly".

Not sure if I'd allow the "present in BUILD_DEPENDS" carve-out; having
ccache in BUILD_DEPENDS won't be enough to make it work (at least with
PORTS_PRIVSEP), as at least CCACHE_DIR still needs to be set. Maybe
go without that at first and revisit if we end up finding something that
really does need it to be present?

On 2026/09/10 04:54, Anthony J. Bentley wrote:
> Klemens Nanni writes:
> > 09.09.2026 13:59, Anthony J. Bentley пишет:
> > > There are probably still several ports with hidden deps on ccache.
> > > This should catch them.
> > > 
> > > If upstream has a knob to turn off ccache, the port should use it.
> >
> > We could also add a few likely candidates to configure args by default,
> > say -DCCACHE=OFF, -CCACHE_Found=FALSE, etc.
> 
> There's no standard name for these though. Currently the only variables
> cmake.port.mk sets are CMAKE_* prefixed variables or language variables
> used by cmake's built-in language modules. It would be confusing for the
> module to try a bunch of random variable names not standardized anywhere
> in hopes of catching the random names individual ports use for ccache.

+1