On Sat 23/01/2021 13:33, Jeremie Courreges-Anglas wrote: > On Fri, Jan 22 2021, Bjorn Ketelaars <[email protected]> wrote: > > On Fri 22/01/2021 08:19, Jeremie Courreges-Anglas wrote: > >> On Fri, Jan 22 2021, Bjorn Ketelaars <[email protected]> wrote: > >> > On Thu 21/01/2021 23:29, Jeremie Courreges-Anglas wrote: > >> >> On Thu, Jan 21 2021, Stuart Henderson <[email protected]> wrote: > >> >> > On 2021/01/21 06:43, Bjorn Ketelaars wrote: > >> >> >> On Wed 20/01/2021 21:42, Jeremie Courreges-Anglas wrote: > >> >> >> > On Wed, Jan 20 2021, Bjorn Ketelaars <[email protected]> wrote: > >> >> >> > > (Maintainer timeout) > >> >> >> > > > >> >> >> > > ocaml is currently a RDEP for all ports that are build with it. > >> >> >> > > This is > >> >> >> > > unnecessary for a couple of them, i.e. net/unison. > >> >> >> > > >> >> >> > As far as I know, programs written in OCaml should run without the > >> >> >> > OCaml > >> >> >> > runtime installed on ${OCAML_NATIVE_ARCHS}, currently i386 and > >> >> >> > amd64. > >> >> >> > Except for OCaml programs that make direct use of the OCaml > >> >> >> > runtime, of > >> >> >> > course... But the rundep is still needed on non-native archs. > >> >> >> > > >> >> >> > > I would like to > >> >> >> > > propose the addition of MODOCAML_RUNDEP and MODOCAML_BUILDDEP, > >> >> >> > > which are > >> >> >> > > modelled on python.port.mk. > >> >> >> > > > >> >> >> > > Default behaviour is not changed. Setting MODOCAML_RUNDEP=No in a > >> >> >> > > Makefile excludes ocaml as RDEP. > >> >> >> > > > >> >> >> > > OK? > >> >> >> > > >> >> >> > Your diff makes sense I think, but you can't just use it to > >> >> >> > disable the run-dep in net/unison: you'd break unison installs on > >> >> >> > non-native archs. > >> >> >> > >> >> >> Oops, I didn't think about the above. > >> >> >> > >> >> >> I will look further for a fitting, clean solution that serves all > >> >> >> arches. > >> >> >> > >> >> > > >> >> > Maybe allow MODOCAML_RUNDEP=yes/no/nonative or some similar thing? > >> >> > >> >> Hmm, so "nonative" would be for ports that produce bytecode-only > >> >> programs, even on native archs? (No idea how many ports are affected.) > >> >> We could call it "bytecode" then, that would seem a bit more descriptive > >> >> than a plain MODOCAML_RUNDEP=Yes in a port Makefile. > >> >> > >> >> And then MODOCAML_RUNDEP=Yes would be useful for ports that always > >> >> use lang/ocaml programs/files at runtime? > >> >> > >> >> > It would be nice to handle the arch detection in ocaml.port.mk rather > >> >> > than in any port using this. > >> >> > >> >> Here's the simple diff I had in mind yesterday, on top of Bjorn's diff. > >> >> It probably helps Bjorn with unison but it would require testing all > >> >> other OCaml programs, and bumping REVISION for a bunch of them. > >> >> Maybe not the easiest or safest approach. > >> >> > >> >> I suspect that most people caring about OCaml in ports have thought > >> >> about this before, hopefully they can share their ideas. :) > >> > > >> > I'm not particular fond of testing all other OCaml programs. > >> > > >> > What do you think of the diff below; lang/ocaml is added to > >> > BUILD_DEPENDS and RUN_DEPENDS unless native-code compilation is possible > >> > *and* the port specifically requests not to. > >> > > >> > lang/ocaml is always added in cases of ! ${OCAML_NATIVE_ARCHS}, and > >> > 'make show=MODOCAML_RUNDEP' would show 'Ignored as native-code > >> > compilation is not supported on this platform'. > >> > > >> > > >> > diff --git ocaml.port.mk ocaml.port.mk > >> > index bda7f1fc763..b5e9c863089 100644 > >> > --- ocaml.port.mk > >> > +++ ocaml.port.mk > >> > @@ -33,8 +33,27 @@ MODOCAML_OCAMLDOC?=ocamldoc > >> > PKG_ARGS+=-Ddynlink=0 > >> > .endif > >> > > >> > -RUN_DEPENDS += lang/ocaml > >> > -BUILD_DEPENDS += lang/ocaml > >> > +MODOCAML_RUN_DEPENDS= lang/ocaml > >> > +MODOCAML_BUILD_DEPENDS= ${MODOCAML_RUN_DEPENDS} > >> > + > >> > +# Add lang/ocaml to BUILD_DEPENDS and RUN_DEPENDS unless native-code > >> > +# compilation is possible (OCAML_NATIVE_ARCHS) and the port > >> > +# specifically requests not to. > >> > +.if ${PROPERTIES:Mocaml_native} > >> > +MODOCAML_BUILDDEP?= Yes > >> > +MODOCAML_RUNDEP?= Yes > >> > +.else > >> > +MODOCAML_BUILDDEP:= Ignored as native-code compilation is not > >> > supported on this platform > >> > >> You still need ocaml at build time, to build the bytecode and other things. > > > > I agree! > > > > It is highly unlikely that one would want to use MODOCAML_BUILDDEP=No in > > a Makefile. I can't think of any reason for offering this knob except > > that it mimics erlang/go/lua/php/python/ruby. > > Unlikely but valid, just think of a port that would use OCaml scripts at > runtime. (You can write OCaml scripts with a shebang line.) > > Anyway I think it's good to allow anything in a module to be disabled, > else people tend to resort to hacks to work around hardcoded choices. > > > Usage of this knob will result in failure early on. BTW if you set > > MODOCAML_BUILDDEP=No in a Makefile *and* build on a non-native arch > > lang/ocaml will still be added to BUILD_DEPENDS. > > Well I misunderstood the intent of your diff, I'll blame it on my coffee > depriveness. :) > > >> > +MODOCAML_RUNDEP:= ${MODOCAML_BUILDDEP} > >> > >> Should still use ?= I think. > > > > Don't think so. If we are building on a non-native arch we want to make > > sure that lang/ocaml is added as BDEP and RDEP, independent how > > MODOCAML_RUNDEP or MODOCAML_BUILDDEP are used in a Makefile. > > [...] > > An OCaml port probably needs lang/ocaml at build time on *all archs*, > I don't see a reason to introduce a difference between native and > non-native archs *at build time*, nor do I see a good reason to ignore > what a port specifies. > > New proposal, still on top of your diff (full diff below). > > | Index: ocaml.port.mk > | =================================================================== > | --- ocaml.port.mk.orig > | +++ ocaml.port.mk > | @@ -37,13 +37,19 @@ MODOCAML_RUN_DEPENDS= lang/ocaml > | MODOCAML_BUILD_DEPENDS= lang/ocaml > | > | # Assume that we want to automatically add ocaml to BUILD_DEPENDS > | -# and RUN_DEPENDS unless the port specifically requests not to. > | +# unless the port specifically requests not to. > | MODOCAML_BUILDDEP?= Yes > | +# Same for RUN_DEPENDS, but MODOCAML_RUN_DEPENDS can take three values: > | +# Yes, No or if-not-native (translates to Yes if native-code is > unsupported) > | MODOCAML_RUNDEP?= Yes > | > | .if ${NO_BUILD:L} == no && ${MODOCAML_BUILDDEP:L} == yes > | BUILD_DEPENDS+= ${MODOCAML_BUILD_DEPENDS} > | .endif > | + > | +.if ${MODOCAML_RUNDEP:L} == if-not-native && ${MODOCAML_NATIVE} == No > | +MODOCAML_RUNDEP = Yes > | +.endif > | .if ${MODOCAML_RUNDEP:L} == yes > | RUN_DEPENDS+= ${MODOCAML_RUN_DEPENDS} > | .endif > > This implements the three choice logic suggested by sthen, doesn't > affect current defaults, and only cares about the runtime dep. > I hope that "if-not-native" is an improvement over "nonative". > > Thoughts?
Behaviour is spot on, and the three choices make sense. Plus, I think that if-not-native is more descriptive than nonative. I would like to see this committed. OK bket@
