#68 pattern match bug involving alternation

[James A Self]

Some pattern match expressions evaluate as false when
they should be true. The examples so far all involve
fairly complex search strings and patterns with
alternation and multiple open ended patterns, such as
1(.e1p,.p).an1"sarcoma". The result varies with the
number of characters inserted before a common search
string that should be matched by a given pattern.

The test results below are from a MUMPS command shell
(part of M2Web) and textchk-dx is the name of a macro
defined below to simplify repeated testing of different
patterns. I believe that in all of the tests below,
@textchk should evaluate as true.

;testing pattern match operator
w $zversion
GT.M V5.0-000 Linux x86

def textchk-dx(textchk) s text=dx f i=1:1:64 w @textchk
w:i#8=0 " " s text=" "_text

s dx="bone (vertebral bodies: t4, t9, t11, t13, l3):
metastatic osteosarcoma"

textchk-dx text?1(.e1p,.p).an1"sarcoma"
01110011 00011000 11001100 11000000 10000000 00110000
11111111 11111111

textchk-dx text?1(.e1p,0p).an1"sarc".e
01110011 00011000 11001100 11000000 10000000 00110000
11111111 11111111

textchk-dx text?1(.e1p,1p).an1"sarc".e
01110011 00011000 11001100 11000000 10000000 00110000
11111111 11111111

textchk-dx text?1(.e1p,.e).an1"sarc".e
11111111 11111111 11111111 11111111 11111111 11111111
11111111 11111111

textchk-dx text?1(.p,.e1p).an1"sarc".e
01110011 00011000 11001100 11000000 10000000 00110000
11111111 11111111

textchk-dx text?1(0p,.e1p).an1"sarc".e
01110011 00011000 11001100 11000000 10000000 00110000
11111111 11111111

textchk-dx text?1(1p,.e1p).an1"sarc".e
01110011 00011000 11001100 11000000 10000000 00110000
11111111 11111111

textchk-dx text?1(.e,.e1p).an1"sarc".e
11111111 11111111 11111111 11111111 11111111 11111111
11111111 11111111

textchk-dx text?1(.e1p).an1"sarc".e
11111111 11111111 11111111 11111111 11111111 11111111
11111111 11111111

textchk-dx text?.1(.e1p).an1"sarc".e
11111111 11111111 11111111 11111111 11111111 11111111
11111111 11111111

textchk-dx text?1(1.e,.e1p).an1"sarc".e
11111111 11111111 11111111 11111111 11111111 11111111
11111111 11111111

textchk-dx text?1(1.p,.e1p).an1"sarc".e
01110011 00011000 11001100 11000000 10000000 00110000
11111111 11111111

textchk-dx text?1(.an,.e1p.an)1"sarc".e
11111111 11111111 11111111 11111111 11111111 11111111
11111111 11111111

[Malleswara Rao P]

Logged In: YES
user_id=220075

Hi James,
Thanks for reporting the bug. GT.M seems to have problems
with pattern expressions mixed with alternations. In order
to find the exact cause, we need much more analysis and
research. We've created the following tracking number to
address this in a future release.

S9F11-002577 Patterns mixed with alternations do not work
as expected

It looks like the problem disappears if you remove the
pattern that follows the alternation by merging into the
alternation such as your last case. For example, you can
use 1(.e1p.an,.p.an)1"sarcoma" instead of 1
(.e1p,.p).an1"sarcoma".

Thanks
-malli
GT.M Developer
Fidelity Information Services, Inc.

[Malleswara Rao P]

Logged In: YES
user_id=220075

Hi James,
Thanks for reporting the bug. GT.M seems to have problems
with pattern expressions mixed with alternations. In order
to find the exact cause, we need much more analysis and
research. We've created the following tracking number to
address this in a future release.

S9F11-002577 Patterns mixed with alternations do not work
as expected

It looks like the problem disappears if you remove the
pattern that follows the alternation by merging into the
alternation such as your last case. For example, you can
use 1(.e1p.an,.p.an)1"sarcoma" instead of 1
(.e1p,.p).an1"sarcoma".

Thanks
-malli
GT.M Developer
Fidelity Information Services, Inc.

[James A Self]

Logged In: YES
user_id=37201

I found another pattern match problem with the repetition
count on an alternation with a single pattern inside.

The first test below should evaluate as true. If a second
alternative is included, it will return true even when the
new alternative fails to match, such as the 7th and 8th
examples below.

;testing pattern match bug
s text="sarcoma"
w text?.1(.e1p)1"sarc".e
0

w text?.2(.e1p)1"sarc".e
0

w text?1(.e1p)1"sarc".e
0

w text?.1(.e1p,0p)1"sarc".e
1

w text?.2(.e1p,0p)1"sarc".e
1

w text?1(.e1p,0p)1"sarc".e
1

w text?.1(.e1p,1p)1"sarc".e
1

w text?.2(.e1p,1p)1"sarc".e
1

w text?1(.e1p,1p)1"sarc".e
0

w text?.1(.e1p,.1p)1"sarc".e
1

w text?.2(.e1p,.1p)1"sarc".e
1

w text?1(.e1p,.1p)1"sarc".e
1

w text?0.1(.e1p,.1p)1"sarc".e
1

w text?0.2(.e1p,.1p)1"sarc".e
1

w text?0.1(.e1p,0p)1"sarc".e
1

w text?0.2(.e1p,0p)1"sarc".e
1

w text?0.1(.e1p)1"sarc".e
0

w text?0.2(.e1p)1"sarc".e
0

[Narayanan Iyer]

Logged In: YES
user_id=97919

Jim,

Thank you for reporting another problem with pattern
matching. After some investigation, we have been able to
identify the issue for both the problems that you reported
in this thread. The fix for both these issues will be
available in the next GT.M release.

Thanks,
Narayanan.

[Narayanan Iyer]

Logged In: YES
user_id=97919
Originator: NO

This issue is fixed as part of S9F11-002577 in GT.M V5.0-000D.