On Sun 2017-07-30 16:53:28 +0200, Stéphane Glondu wrote: > sks FTBFS with OCaml 4.05.0. Relevant log: >> [...] >> ocamlc -I lib -I bdb -I +zarith -I +cryptokit -ccopt -g -ccopt -O2 >> -ccopt -fdebug-prefix-map=/tmp/sks-1.1.6=. -ccopt -fstack-protector-strong >> -ccopt -Wformat -ccopt -Werror=format-security -ccopt -O3 -ccopt >> -Werror-implicit-funct >> File "eventloop.ml", line 133, characters 15-19: >> Error: This expression has type ?cloexec:bool -> Unix.file_descr >> but an expression was expected of type >> Unix.file_descr = Unix.file_descr >> ocamlc -I lib -I bdb -I +zarith -I +cryptokit -ccopt -g -ccopt -O2 >> -ccopt -fdebug-prefix-map=/tmp/sks-1.1.6=. -ccopt -fstack-protector-strong >> -ccopt -Wformat -ccopt -Werror=format-security -ccopt -O3 -ccopt >> -Werror-implicit-funct >> Makefile:357: recipe for target 'eventloop.cmo' failed >> make[2]: *** [eventloop.cmo] Error 2 >> [...]
thanks for the heads-up, Stéphane!
> Attached is a patch. Unfortunately it is not compatible with the
> version of OCaml currently in unstable, so you'll have to wait until
> we upload ocaml >= 4.05.0.
and thanks for the patch.
But the fact that we need it is pretty disappointing, and it's not the
first time that a backwards-incompatible API change was introduced in
what looks like a minor release of OCaml. (see also the bytes changes
introduced in 4.02 or 4.03 for example).
What does OCaml upstream think that packages that are actually supposed
to ship against multiple architectures and versions of the ocaml
compiler should do in these cases? if we include this patch in debian,
we will have to drop it in backports, etc. This kind of bookkeeping is
annoying and error-prone. If it happened at major version transitions
for OCaml, i'd be more inclined to put up with it, but when it's on
seemingly minor updates, it makes me grumpy :/
--dkg
signature.asc
Description: PGP signature
