OSX Compatibility

[rgrinberg]

Running jenga on OSX against of the examples gives:

 ("unhandled exception"
   ((lib/monitor.ml.Error_
     ((exn
       (Unix.Unix_error "Invalid argument" select
        "((read (-1)) (write ()) (except ()) (timeout Never))"))
      (backtrace
       ("Raised at file \"lib/core_unix.ml\", line 48, characters 11-42"
        "Called from file \"lib/async_inotify.ml\", line 98, characters 10-77"
        "Called from file \"lib/result.ml\", line 104, characters 10-14"
        "Re-raised at file \"lib/result.ml\", line 111, characters 23-26"
        "Called from file \"lib/deferred.ml\", line 12, characters 64-67"
        "Called from file \"lib/jobs.ml\", line 214, characters 10-13" ""))
      (monitor
       (((name try_with) (here ()) (id 4) (has_seen_error true)
         (someone_is_listening true) (kill_index 0))))))
    (Pid 57794)))

Seems like this is because inotify is absent from OSX.

[samoht]

Any progress on this ? I've just been bitten by the same issue.

[ghost]

Not that I know of, I'll ping people involved.

[ghost]

The plan is to replace inotify by a portable library, but the work hasn't started yet.

[samoht]

Would be nice to at least not fail on inotify when not running in a polling mode.

Also there are some "scary" warnings which would be nice to not print at the start of the day:

$ jenga
cant read: /proc/cpuinfo
no processor lines seen in: /proc/cpuinfo
[...]

[avsm]

A fully portable inotify is kind of tricky -- being able to disable the polling functionality entirely and do a one-shot build would be very useful.

[samoht]

I managed to get past this first error message but I then get a new one:

$ jenga -no-notifiers
cant read: /proc/cpuinfo
no processor lines seen in: /proc/cpuinfo
*** jenga: [60237] root=/Users/thomas/git/jenga/examples/simple-ocaml-build, j=1, f=1
*** jenga: ERROR: jengaroot.ml: unable to digest file
*** jenga: exited with code 254
*** jenga: 0/2 targets are up to date
*** jenga: failed (#1, 0ms, heap(Kb)= 3721/5672, major=2, stat=1, digest=1, ls=0, db-save=0)
*** jenga: todo: 0 (0 / 2) !1 ~1 j=0+0 con=0 act=0, finished
*** jenga: ERROR: (summary) jengaroot.ml: unable to digest file
*** jenga: exited with code 254
*** jenga: build finished; not in polling mode so quitting
*** jenga: Quit; exit_code = 1

[mshinwell]

Set the environment variable JENGA_USE_OCAML_DIGEST and try again.

We should fix jenga so that it doesn't have a hard-coded dependency on /usr/bin/md5sum (which of course is called "md5" on the Mac).

[mshinwell]

Also, it might be helpful to run with -verbose.

[samoht]

seems to work now, thanks!

$ export JENGA_USE_OCAML_DIGEST=1
$ jenga -no-notifiers -verbose
using internal ocaml digest instead of external call to md5sum
cant read: /proc/cpuinfo
no processor lines seen in: /proc/cpuinfo
*** jenga: [60913] root=/Users/thomas/git/jenga/examples/simple-ocaml-build, j=1, f=1
- build . jengaroot.ml
*** jenga: reading jengaroot.ml
*** jenga: finished reading jengaroot.ml (1.032s)
- build hello stdout
+ bash -c 'ocamldep -native hello.ml'
hello.cmo : world.cmi
hello.cmx : world.cmx
- exit hello stdout, 26ms, code 0
- build hello stdout
+ bash -c 'ocamldep -native world.ml'
world.cmo : world.cmi
world.cmx : world.cmi
- exit hello stdout, 7ms, code 0
- build hello stdout
+ bash -c 'ocamldep -native world.mli'
world.cmi :
- exit hello stdout, 8ms, code 0
- build hello world.cmi
+ bash -c 'ocamlopt -c world.mli'
- exit hello world.cmi, 20ms, code 0
- build hello world.cmx
+ bash -c 'ocamlopt -c world.ml'
- exit hello world.cmx, 29ms, code 0
- build hello hello.cmi
+ bash -c 'ocamlopt -c hello.ml'
- exit hello hello.cmi, 32ms, code 0
- build hello hello.exe
+ bash -c 'ocamlopt.opt world.cmx hello.cmx -o hello.exe'
- exit hello hello.exe, 66ms, code 0
*** jenga: 14/14 targets are up to date
*** jenga: done (#1, 207ms, heap(Kb)= 20792/210472, major=2, stat=133, digest=12, ls=10, db-save=0) -- HURRAH
*** jenga: todo: 0 (14 / 14) j=0+0 con=14 act=7, finished
*** jenga: build finished; not in polling mode so quitting
*** jenga: Quit; exit_code = 0

[mshinwell]

marvellous.

[avsm]

awesome!

On 11 Jun 2014, at 16:54, Mark Shinwell notifications@github.com wrote:

marvellous.

—
Reply to this email directly or view it on GitHub.

[Nick-Chapman]

The "scary" warnings:

$ jenga
cant read: /proc/cpuinfo
no processor lines seen in: /proc/cpuinfo
*** jenga: [60913] root=/Users/thomas/git/jenga/examples/simple-ocaml-build, j=1, f=1

are because jenga attempts to determine the number of CPUs in order to choose a sensible default for the -j and -f flags. It fails, so you get j-1, f=1 (as it shows you).

If you want parallel builds, you will have to set -j on the command line. (#cpus-1) is a reasonable choice - but please experiment. Sadly this wont make the scary message go away.

The proper fix is for jenga to use an architecture independent way of determining the CPU count.

[mshinwell]

We're fixing Core to use the sysconf method to determine the number of cores, which will fix the /proc/cpuinfo problems (and make it know how many cores there are on a Mac).

[avsm]

Need to guard sysconf around Linux|Darwin, as BSDs use sysctl to retrieve this information.

On 12 Jun 2014, at 10:47, Mark Shinwell notifications@github.com wrote:

We're fixing Core to use the sysconf method to determine the number of cores, which will fix the /proc/cpuinfo problems.

—
Reply to this email directly or view it on GitHub.

[mshinwell]

Really? I thought sysconf was a POSIX standard function...

[avsm]

sysconf is, but _SC_NPROCESSORS_ONLN isnt mandated by POSIX yet [1]. I've just checked on modern OpenBSD and it does have it now, so it may be ok to depend on. Still good to guard it behind a #ifdef and default to one if not present.

[1] http://austingroupbugs.net/view.php?id=339

-a

On 16 Jun 2014, at 09:10, Mark Shinwell notifications@github.com wrote:

Really? I thought sysconf was a POSIX standard function...

—
Reply to this email directly or view it on GitHub.

[bsansouci]

Hey everyone, I'm trying to use jenga built from master on OS X 10.11 beta and am getting this error when running the command jenga -prog -P example/.runtest

*** jenga: [63054] root=/Users/benjamin/Downloads/jengaroot, sys=Darwin, j=1, f=1 (NO_VERSION_UTIL)
*** jenga: ERROR: cannot take any lock in "/Users/benjamin/Downloads/jengaroot/.jenga" (running on nfs?)
*** jenga: Quit; exit_code = 4

I initially thought it was related to flock which apparently shouldn't be used on OS X (and lockf is preferred). I tried removing the flock from core and building jenga with that local version but got the same error. I'm a little bit at a loss now...
I'll continue commenting here if I can find anything, but I'd love some help if anyone has any idea.

[ghost]

@bsansouci, I noticed this as well the other day when trying jenga on my laptop. IIRC replacing flocked && lockfed by flocked || lockfed in core gave me a working jenga. Could you try that? I'll get this done properly for the next release

[bsansouci]

hey @diml thanks for that! I was building core wrong and it wasn't picking up my changes. This worked but I don't think we need to use both flock and lockf seeing how flock is supposed to be strictly better than lockf.
I'm now getting a bunch of preprocessing errors but those are due to my own package not jenga. Thanks a lot for your help :)

[ghost]

No problem. We should include the proposed fix in a minor release of core, and we'll look for a better one for future releases; we'll need to add more #if variables to disable flock only on OSX