Failing test on OSX

[hoosierEE]

Here's what I get:

$ tup
[ tup ] [0.000s] Scanning filesystem...
[ tup ] [0.039s] Reading in new environment variables...
[ tup ] [0.039s] No Tupfiles to parse.
[ tup ] [0.039s] No files to delete.
[ tup ] [0.049s] Executing Commands...
* 1) TEST test/core/g110.ijs                                                                                           
test/core/g110.ijs: failed
|NaN error: sgn
|       test x=:o._10+?5 10$20
|[-8] test/core/g110.ijs

 *** tup errors ***
 *** Command ID=1401 failed with return value 1
tup error: Expected to write to file 'build/test/g110.ijs.pass' from cmd 1401 but didn't
 *** Additionally, command 1401 failed to process input dependencies. These should probably be fixed before addressing the command failure.
 2) [0.042s] TEST test/core/g111a.ijs                                                                                  
 3) [0.051s] TEST test/core/g110s.ijs                                                                                  
 4) [0.052s] TEST test/core/g111.ijs                                                                                   
 [                                        ETA~=5s  Remaining=312 Active=0                                        ]   1%
 *** tup: 1 job failed.

I examined the test more closely with unbox/bin/jconsole, and it appears to fail when there are zeros present in the x table, but it succeeds when there are no zeros.

[iocane]

Can you start J console and try

%0

do you get NaN error or _?

[hoosierEE]

I get _

[iocane]

Looks like a problem with handling overflow. Try this,

put this file:
http://www.jsoftware.com/gitlist/jsource.git/blob/master/asm/macasm64.o
in src/libj/macosx.

Remove -DNOASM from lib_cflags in config/macosx

Change:
#define C_NA 1
to
#define C_NA 0
in
src/libj/js.h

In config/macosx, change:
: {libj} |> ^ LINK %f^ $(CC) %f -o %o $(lib_lflags) |> $(libj)
to
: {libj} |> ^ LINK %f^ $(CC) %f src/libj/macosx/macasm.o -o %o $(lib_lflags) |> $(libj)

and try again.

[hoosierEE]

Shiny new errors after the suggested changes:

* 132) TEST test/sparse/gsp130.ijs                                                                                     
test/sparse/gsp130.ijs: failed
|NaN error: f
|   (p%q)-:((2;x)$.p)    %(2;y)$.q

 *** tup errors ***
 *** Command ID=1969 failed with return value 1
tup error: Expected to write to file 'build/test/gsp130.ijs.pass' from cmd 1969 but didn't
 *** Additionally, command 1969 failed to process input dependencies. These should probably be fixed before addressing the command failure.
 133) [0.061s] TEST test/sparse/gsp122.ijs                                                                             
 134) [0.291s] TEST test/sparse/gsp120.ijs                                                                             
 [                                        ETA~=51s Remaining=374 Active=1                                        ]  26%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1993 failed: bin/jconsole -test test/sparse build/test gsp420.ijs
 [                                        ETA~=55s Remaining=374 Active=0                                        ]  26%
 *** tup: 2 jobs failed.

%0 still returns _

[iocane]

%0 = _ is correct. I wanted to make sure it didn't return NaN, that would be really bad :P)

Did the other error actually go away? Make sure with

bin/jconsole -debug test/core/g110.ijs

[hoosierEE]

Ah, no it did not:

$ bin/jconsole -debug test/core/g110.ijs
   NB. *y ------------------------------------------------------------------

   qct  =: 2^_44
   sgn  =: (%|) * >:!.0&qct@|
   test =: sgn -: *

   test x=:1=?2 5$2
1
   test x=:_10+?13$20
1
   test x=:o._10+?5 10$20
|NaN error: sgn
|       test x=:o._10+?5 10$20
|[-8] test/core/g110.ijs

[iocane]

Actually, this isn't related to overflow.

Can you paste x, also, what does this give >:!.0&qct@| x=:o._10+?5 10$20?

[hoosierEE]

On Wed, Jan 20, 2016 at 3:55 PM, Thomas Costigliola <
notifications@github.com> wrote:

:!.0&qct@| x=:o._10+?5 10$20

:!.0&qct@| x=:o._10+?5 10$20

1 1 1 1 0 1 1 1 1 1

0 1 1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1 1 1

0 1 1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1 1 1

x

_31.4159 _18.8496 _3.14159 12.5664 0 6.28319 25.1327 _25.1327
6.28319 12.5664

   0 _18.8496   15.708   15.708 _9.42478 _6.28319 _21.9911 _12.5664

_31.4159 28.2743

_25.1327 9.42478 3.14159 15.708 _28.2743 _9.42478 21.9911 9.42478
_21.9911 _9.42478

   0  9.42478  6.28319  _15.708 _6.28319  _15.708  3.14159 _28.2743

15.708 _18.8496

_21.9911 9.42478 _12.5664 _18.8496 _3.14159 9.42478 _15.708 _28.2743
_21.9911 25.1327

Failling on zeros still.

[iocane]

Okay, so the NaN error is coming from (%|) x or the multiplication. Can you try (%|) x? If so try (%|) 0.

Also, can you try using a debug build. Set CONFIG_RELEASE=n in tup.conifg.

[hoosierEE]

The minimal failing case is a list with both a zero and a floating-point number:

   (%|)0 1.1
|NaN error
|       (%|)0 1.1

Applying it item-by-item seems to be alright. Suspecting some sort of boolean optimization getting applied when it shouldn't.

   (%|)each 0 1.1
+-+-+
|0|1|
+-+-+
   (%|)"0]0 1.1
0 1

[iocane]

This is going to be difficult for me without a debugger.

What are the results of

 0 % 0
 0 1.1 % 0 1.1
 (0.1 - 0.1) % (0.1 - 0.1)

[hoosierEE]

This bug seems similar to the E. bug having trouble with boolean (or was it integer?) arguments, which I recall seeing in the mailing list a month or two ago, but can't find now.

   0 % 0
0
   0 1.1 % 0 1.1
|NaN error
|   0 1.1    %0 1.1
   (0.1 - 0.1) % (0.1 - 0.1)
0

   1.1 0 %/ 1 1 0 ,. 1 0.1 1  NB. other tests....
1.1 1.1
1.1  11
  _ 1.1

  0   0
  0   0
  0   0
   1.1 0%/1 0 0,.1 0.1 1
|NaN error
|   1.1 0    %/1 0 0,.1 0.1 1
   1.1 0 %/ 1 0.8 0 ,. 1 0.1 1
  1.1 1.1
1.375  11
    _ 1.1

    0   0
    0   0
    0   0
   1.1 0 %/ 1 0 0 ,. 1 0.1 1
|NaN error
|   1.1 0    %/1 0 0,.1 0.1 1
   1.1 0 %/ 1 0 1 ,. 1 0.1 1
|NaN error
|   1.1 0    %/1 0 1,.1 0.1 1
      1.1 0 %/ 0 0 1 ,. 1 0.1 1
|NaN error
|   1.1 0    %/0 0 1,.1 0.1 1
   1.1 0 %/ 0 0 0 ,. 1 0.1 1
|NaN error
|   1.1 0    %/0 0 0,.1 0.1 1
   1.1 0 %/ 0 0 0 ,. 0 0 0
_ _
_ _
_ _

0 0
0 0
0 0

[iocane]

0 1.1 % 0 1.1 Both sides are float because of the 1.1 so there isn't any conversions boolean or otherwise. You can confirm with 3!:0 ] 0 1.1. Can you use your build script updates with the original commit to check that the J software merge didn't mess up any system related settings?

Also, does this happen when you set CONFIG_CLANG=y and CONFIG_CLANG=n?

[hoosierEE]

Yep, 3!:0]0 1.1 returns 8.

$ cat tup.config 
CONFIG_RELEASE=y
CONFIG_TEST=y
CONFIG_X86_64=y
CONFIG_READLINE=y
CONFIG_CLANG=n

$ tup
[ tup ] [0.000s] Scanning filesystem...
[ tup ] [0.009s] Reading in new environment variables...
[ tup ] [0.009s] No Tupfiles to parse.
[ tup ] [0.009s] No files to delete.
[ tup ] [0.014s] Executing Commands...
* 1) TEST test/core/g110.ijs                                                                                           
test/core/g110.ijs: failed
|NaN error: sgn
|       test x=:o._10+?5 10$20
|[-8] test/core/g110.ijs

 *** tup errors ***
 *** Command ID=1401 failed with return value 1
tup error: Expected to write to file 'build/test/g110.ijs.pass' from cmd 1401 but didn't
 *** Additionally, command 1401 failed to process input dependencies. These should probably be fixed before addressing the command failure.
 [                                        ETA~=7s  Remaining=261 Active=3                                        ]   0%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1407 failed: bin/jconsole -test test/core build/test g110p.ijs
 [                                        ETA~=8m  Remaining=261 Active=2                                        ]   0%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1405 failed: bin/jconsole -test test/core build/test g110i.ijs
 [                                        ETA~=8m  Remaining=261 Active=1                                        ]   0%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1409 failed: bin/jconsole -test test/core build/test g110s.ijs
 [                                        ETA~=8m  Remaining=261 Active=0                                        ]   0%
 *** tup: 4 jobs failed.

$ cat tup.config 
CONFIG_RELEASE=y
CONFIG_TEST=y
CONFIG_X86_64=y
CONFIG_READLINE=y
CONFIG_CLANG=y

$ tup
[ tup ] [0.000s] Scanning filesystem...
[ tup ] [0.007s] Reading in new environment variables...
[ tup ] [0.007s] No Tupfiles to parse.
[ tup ] [0.007s] No files to delete.
[ tup ] [0.013s] Executing Commands...
 [                                        ETA~=??? Remaining=262 Active=4                                        ]   0%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1407 failed: bin/jconsole -test test/core build/test g110p.ijs
 [                                        ETA~=??? Remaining=262 Active=3                                        ]   0%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1409 failed: bin/jconsole -test test/core build/test g110s.ijs
 [                                        ETA~=??? Remaining=262 Active=2                                        ]   0%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1401 failed: bin/jconsole -test test/core build/test g110.ijs
 [                                        ETA~=??? Remaining=262 Active=1                                        ]   0%tup error: FUSE did not appear to release all file descriptors after the sub-process closed.
 *** Command ID=1405 failed: bin/jconsole -test test/core build/test g110i.ijs
 [                                        ETA~=??? Remaining=262 Active=0                                        ]   0%
 *** tup: 4 jobs failed.