converge doesn't work right

[bakul]

(1 2 1)\1 yields 1 2 in k but 1 in kona

(1 2 1)\0 runs out of eval size limit in k but just yields 0 in kona

[tavmem]

I agree with your observations on the behavior differences between k2.8 and kona.

However, I don't see any documentation of "converge" in the K 2.0 reference manual. The index has the following entries for \

\
  Abort  171
  command  198, 200
  Scan  138
  Scan Dyad  137
  Stop / Trace  175

None of the pages listed appear to address the the "converge" capability.

Was this a new feature introduced after K 2.0? ... or an undocumented feature of k?

[tavmem]

Well ... whether documented or not ... Shakti k (aka k7) responds in a manner similar to k2.8

$ rlwrap -n ./k
2019-05-24 16:37:43 4core 1gb avx2 © shakti l2.0 test
 (1 2 1)\1
1 2
 (1 2 1)\0
ws
$ 

[tavmem]

Still, it would still be good to have an better understanding of what "converge" is attempting to do, in a more general case than two examples (one of which fails).

BTW: the index in the K 2.0 reference manual is not complete. At a minimum, it leaves out "scan monad"

\
  Abort  171
  command  198, 200
  Scan  138
  Scan Dyad  137
  Stop / Trace  175
  Scan Monad  139

[tavmem]

A second omission in the K 2.0 Reference Manual index ("Each Left"):

\
  Abort  171
  command  198, 200
  Each Left  125
  Scan  138
  Scan Dyad  137
  Scan Monad 139
  Stop / Trace  175

[bakul]

It is basically scan monad, documented in the reference manual. It is related to over monad similar to how scan dyad is related to over dyad_. The over monad iterates, while the scan monad traces the iteration.

kona handles the over monad correctly for list/x and f/x and the scan monad for f\x but not for list\x.

Incidentally typing \' to k3 shows you the CONVERGE operation:

...
m/ CONVERGE     n m/ DO         b m/ WHILE
m\ CONVERGE     n m\ DO         b m\ WHILE
...

where m is any monad, including a list.

[tavmem]

Thanks !! I think my problem was conceptualizing a list as completely different from a monadic function.

Scan Monad is defined as being one of the following

 f\ x 
 n f\ x 
 b f\ x

where f must be a monadic function.

[tavmem]

To (hopefully) demonstrate the problem a bit clearer using the sample monadic function from the Scan Monad section (page 139) of the K 2.0 reference manual:

kona      \ for help. \\ to exit.

  f:{:[x ! 2; x; _ x % 2]}

  f/ 12
3
  f\ 12
12 6 3

Kona works fine using the function. Kona should work exactly the same using the list that is equivalent to that function over the relevant function domain:

  f'!13
0 1 1 3 2 5 3 7 4 9 5 11 6

  0 1 1 3 2 5 3 7 4 9 5 11 6/ 12
3
  0 1 1 3 2 5 3 7 4 9 5 11 6\ 12
12

but it does not.

[tavmem]

This is a regression. It was broken by commit a5a1903c95d8d4b79fe2941864b403cb25387d80 on Feb 3, 2015.

[tavmem]

Removing the Feb 3, 2015 "fix" is easy ... but then 6 tests fail:

Failed. These are not equal:
2 , m:3 4#!12; m\2
********************************
2
--------------------------------

Failed. These are not equal:
6 , m:3 4#!12; 1 m\2
********************************
6
--------------------------------
(2
 8 9 10 11)

Failed. These are not equal:
(,"alpha") , " "\"alpha"
********************************
,"alpha"
--------------------------------

Failed. These are not equal:
("";"lph";"") , "a"\"alpha"
********************************
(""
 "lph"
 "")
--------------------------------

Failed. These are not equal:
(,"a";"pha") , "l"\"alpha"
********************************
(,"a"
 "pha")
--------------------------------

Failed. These are not equal:
("";"";"";"") , "a"\"aaa"
********************************
(""
 ""
 ""
 "")
--------------------------------

The Feb 3, 2015 "fix" forced these cases to use "scanDyad" instead of "scanMonad". While that worked for these 6 cases, there does not appear to be a Dyad in any of the 6.

What corresponds to the "f" argument is either a matrix (list of lists), or an atom ... so, all 6 seem to be "scanMonad" cases.

This indicates that there is something wrong with "scanMonad" that should be fixed.

[tavmem]

The commit above is not what I consider an optimal fix in that it still uses "scanDyad" to process some commands where "f" is clealy a monad. However, all tests pass and it's an easy fix.

kona      \ for help. \\ to exit

  (1 2 1)\1
1 2

  0 1 1 3 2 5 3 7 4 9 5 11 6\ 12
12 6 3

  (1 2 1)\0

In the final case, Kona runs out of space and freezes. Not elegant, but Kona is no longer responding with an incorrect result.