Can't install on OS X with latest release

[yminsky]

It's not obvious to me what the issue is here, but here's the error I get from opam. Does that mean that it can't find the C compiler? I'm a little confused.

=-=- Processing actions -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=  🐫
[ERROR] The compilation of re2 failed at "jbuilder build -p re2 -j 6".

#=== ERROR while installing re2.v0.9.1 ========================================#
# opam-version 1.2.2
# os           darwin
# command      jbuilder build -p re2 -j 6
# path         /Users/yminsky/.opam/4.06.0/build/re2.v0.9.1
# compiler     4.06.0
# exit-code    1
# env-file     /Users/yminsky/.opam/4.06.0/build/re2.v0.9.1/re2-88727-1872ec.env
# stdout-file  /Users/yminsky/.opam/4.06.0/build/re2.v0.9.1/re2-88727-1872ec.out
# stderr-file  /Users/yminsky/.opam/4.06.0/build/re2.v0.9.1/re2-88727-1872ec.err
### stderr ###
# Error: Program cc -O2 -pipe not found in PATH (context: default)

[xclerc]

(For the record, I am not able to reproduce the problem on an old 10.11.x macOS.)

[Chris00]

It look like it is trying to find the program cc -O2 -pipe — which of course fails. The full log would be useful to pinpoint the action that is being executed. PR ocaml/dune#336 may help to solve that. (Discussion with @rgrinberg)

[yminsky]

@xclerc I'm running on a fully up-to-date mac, FWIW.

@rgrinberg, should I just wait for that feature to land, and then try again?

[Chris00]

re2 may need to be adapted. Running in verbose mode and uploading the log here would help.

[xclerc]

(Another data point: unable to reproduce on a 10.13.x macOS, with the latest Xcode 9.{0,2}.x)

[xclerc]

@Chris00 while your explanation is more than plausible, when grepping the
latest jbuilder master, it seems that the only calls that could produce the message
are in bootstrap.ml and are related to ocamlc/ocamllex/ocamldep. (But I am
quite possibly not using the correct jbuilder version.)

[Chris00]

@xclerc I agree — quickly looking at the sources of jbuilder, this seems to be the only place. However, bootstrap.ml is only used when compiling jbuilder itself, it should not play any role when compiling other libraries.

[xclerc]

Indeed, and it just adds to my confusion...

[yminsky]

Urk. Infuriating. I did an opam update, and now the problem is gone. So I don't know what happened, but I might as well close this one.

[dcastrop]

I have the exact same issue on a fully up-to-date mac, when calling opam install re2.v0.9.1. Opam update did not solve the problem for me, so I had to switch to the ocaml version 4.05.0. I briefly tried to understand what was going on with 4.06.0, and it seems like an issue in jbuilder, but I didn't get too far. I write below a few comments, in case somebody wants to take a further look (question: is this really an issue in the jbuilder repo, or is this me misunderstanding something?).

---

• The result of ocamlc -config with OCaml 4.05.0 contains no c_compiler field. However, using 4.06.0, it contains c_compiler: cc -O2 -pipe.

• In the code of jbuilder:

  • Function Context.create returns a context with c_compiler set to one of the following: a) the value of that field of the output of ocamlc -config, if it exists; b) the first word of the field "bytecomp_c_compiler" (i.e. the substring until the first space), otherwise.
  • Function Gen_rules.gen calls SC.resolve_program on ctx.c_compiler. Function resolve_program treats its input as a path to a binary, and if it is not found, this results on a Run (Error e, _) action. For 4.06.0, the input to resolve_program is cc -O2 -pipe, which obviously fails (as @Chris00 said).
  • Compiling jbuilder with the patch below seems to fix the issue (at least for me), but I doubt it is the correct way (e.g. it ignores the flags in -O2 -pipe).

diff --git a/src/context.ml b/src/context.ml
index bc0383f..6cbf3a8 100644
--- a/src/context.ml
+++ b/src/context.ml
@@ -287,25 +287,26 @@ let create ~(kind : Kind.t) ~path ~base_env ~env_extra ~name ~merlin ~use_findli
     else
       env,env_extra
   in
+  let split_prog s =
+    let len = String.length s in
+    let rec loop i =
+      if i = len then
+        (s, "")
+      else
+        match s.[i] with
+        | ' ' | '\t' ->
+          (String.sub s ~pos:0 ~len:i,
+           String.sub s ~pos:i ~len:(len - i))
+        | _ -> loop (i + 1)
+    in
+    loop 0
+  in
   let c_compiler, ocamlc_cflags, ocamlopt_cflags =
     match get_opt "c_compiler" with
     | Some c_compiler -> (* >= 4.06 *)
+      let c_compiler, _ = split_prog c_compiler in
       (c_compiler, get "ocamlc_cflags", get "ocamlopt_cflags")
     | None ->
-      let split_prog s =
-        let len = String.length s in
-        let rec loop i =
-          if i = len then
-            (s, "")
-          else
-            match s.[i] with
-            | ' ' | '\t' ->
-              (String.sub s ~pos:0 ~len:i,
-               String.sub s ~pos:i ~len:(len - i))
-            | _ -> loop (i + 1)
-        in
-        loop 0
-      in
       let c_compiler, ocamlc_cflags = split_prog (get "bytecomp_c_compiler") in
       let _, ocamlopt_cflags = split_prog (get "native_c_compiler") in
       (c_compiler, ocamlc_cflags, ocamlopt_cflags)

[yminsky]

This has reemerged for me now that I'm trying out 4.06 again. Rudi asked me for more debugging information, so here it is:

incr_dom $ ocamlc -config | grep ‘compiler:’
c_compiler: cc -O2 -pipe
bytecomp_c_compiler: cc -O2 -pipe -O2 -fno-strict-aliasing -fwrapv  -D_FILE_OFFSET_BITS=64 -D_REENTRANT
native_c_compiler: cc -O2 -pipe -O2 -fno-strict-aliasing -fwrapv -D_FILE_OFFSET_BITS=64 -D_REENTRANT

Given that this is apparently a jbuilder issue, I'll mint one, pointing to this issue.

[yminsky]

See: ocaml/dune#392

[smolkaj]

Same problem, also MacOS. Builds with OCaml 4.05, fails to build with OCaml 4.06.

[rgrinberg]

should be fixed in jbuilder#master.

[smolkaj]

Yes, indeed. Thanks.