Possible loop in multiple tests

[dougburns]

Several tests run for a long time consuming many minutes of CPU.
Without intervention the build eventually times out, so this may be a hard loop.
Killing the problem tests allows the build to complete with failures.

test-canonicalize
test-mbscasestr1
test-mbsstr1
test-unsetenv

[dougburns]

Stack traces:

test-canonicalize

  Traceback:                                                                                                                    
    DSA      Entry       E  Offset  Statement   Load Mod             Program Unit                   Service  Status             
                                                                                                                                
    1        hash_lookup -00000574              *PATHNAM                                                     Call               
    2        seen_file   +0000004A              *PATHNAM                                                     Call               
    3        canonicalize_filename_mode_stk                                                                                     
                         +0000058E              *PATHNAM                                                     Call               
    4        canonicalize_file_name                                                                                             
                         +00000014              *PATHNAM                                                     Call               
    5        main        +000001EE              *PATHNAM                                                     Call               
    6        CELQINIT    +00001ACA              CELQLIB              CELQINIT                       HLE77C0  Call               

test-mbscasestr1

        Traceback:                                                                                                                     
      DSA      Entry       E  Offset  Statement   Load Mod             Program Unit                   Service  Status              
                                                                                                                                   
      1        mbscasestr  +00000A4E              *PATHNAM                                                     Call                
      2        main        +0000055E              *PATHNAM                                                     Call                
      3        CELQINIT    +00001ACA              CELQLIB              CELQINIT                       HLE77C0  Call                

test-mbsstr1

      Traceback:                                                                                                                     
  DSA      Entry       E  Offset  Statement   Load Mod             Program Unit                   Service  Status              
                                                                                                                               
  1        mbsstr      +00000702              *PATHNAM                                                     Call                
  2        main        +00000560              *PATHNAM                                                     Call                
  3        CELQINIT    +00001ACA              CELQLIB              CELQINIT                       HLE77C0  Call                

test-unsetenv

      Traceback:                                                                                                                    
    DSA      Entry       E  Offset  Statement   Load Mod             Program Unit                   Service  Status             
                                                                                                                                
    1        CEL4ENVA    +00000A68              CELQLIB              CEL4ENVA                       HLE77C0  Call               
    2        __setenv_a  +000006F2              CELQLIB                                             UI75815  Call               
    3        __unsetenv_a+0000002A              CELQLIB                                             UI75815  Call               
    4        rpl_unsetenv+000000E6              *PATHNAM                                                     Call               
    5        main        +00000150              *PATHNAM                                                     Call               
    6        CELQINIT    +00001ACA              CELQLIB              CELQINIT                       HLE77C0  Call               

No pattern immediately stands out. I'll generate listings for the modules and dig deeper into each case.

[dougburns]

test-canonicalize loops trying to process "/etc/passwd".
The problem is caused by /etc being a symbolic link that starts with a sysplex symbol ($SYSNAME).
canonicalize will need changes to deal with these sysplex symbolic links.
I've added a temporary change to avoid the loop by detecting a sysplex symbol and return an error.
Slightly oddly, this causes the test to pass - the "/etc/passwd" check doesn't fail the test because it doesn't exist on all systems.
I've raised issue #13 to look at a proper fix.

[dougburns]

test-mbscasestr1 isn't looping, it just runs very inefficiently.
The problem occurs in the sections with the comment:

/* Check that the asymptotic worst-case complexity is not quadratic.  */

This is supposed to ensure that a complex comparison uses the more efficient Knuth-Morris-Pratt algorithm.
The selection criteria for that function aren't ever met, so the test is slogging its way through the search using the more inefficient brute-force approach.
The fast algorithm isn't used because strnlen is returning an incorrect value for a string length.
This is an issue in zoslib - ibmruntimes/zoslib#18

[dougburns]

test-mbsstr1 is almost identical to test-mbscasestr1, so presumably is performing in the same way.
The strnlen fix mentioned above causes test-mbsstr1 to complete in a timely manner too.

[dougburns]

test-unsetenv loops because the z/OS implementation of getenv/unsetenv provide inconsistent handling of an envar string that is altered after it has been set.
This same problem has been seen in gettextport (also in the embedded gnulib code).
See: zopencommunity/gettextport#15 and https://github.com/ZOSOpenTools/gettextport/pull/14/files
We'll need to patch the text to prevent the loop until the underlying issue is resolved.

[dougburns]

fixed by #11