all devices regression report - 2017-11-20-08:10:34 - 1b64014143a8a571e5dbff3346c6ad902603dda1

Differences in all devices errors using gs/examples files

Previous Revision: cea5d8a72dab953fe84ab8f7280003d377559fff
 Current Revision: 1b64014143a8a571e5dbff3346c6ad902603dda1

commit 1b64014143a8a571e5dbff3346c6ad902603dda1
Author:     Robin Watts <[email protected]>
AuthorDate: Mon Nov 20 14:33:03 2017 +0000
CommitDate: Mon Nov 20 14:38:01 2017 +0000

    Fix release/debug difference.
    
    In searching for the indeterminism in 11-14.PS, I found that I could
    run the same command line in release and debug builds on peeved, and
    get different results:
    
     touch out1.pgm ref1.pgm && rm out*.pgm ref*.pgm &&
     bin/gs -sOutputFile=ref%d.pgm -dMaxBitmap=10000 -sDEVICE=pgmraw -r300
     -Z: -sDEFAULTPAPERSIZE=letter -dNOPAUSE -dBATCH -K2000000 -dClusterJob
     -dJOBSERVER %rom%Resource/Init/gs_cet.ps cutdown >& ~/log &&
     debugbin/gs -sOutputFile=out%d.pgm -dMaxBitmap=10000 -sDEVICE=pgmraw
     -r300 -Z: -sDEFAULTPAPERSIZE=letter -dNOPAUSE -dBATCH -K2000000
     -dClusterJob -dJOBSERVER %rom%Resource/Init/gs_cet.ps cutdown
     >& ~/log2 && md5sum ref*.pgm out*.pgm && less ~/log ~/log2
    
    After cutting down the file as much as I could, Chris reduced it
    simply to:
    
     3E9 3E9 moveto 1 1 lineto stroke showpage
    
    Thanks for this!
    
    Tracing through the code, I discovered that the difference came down
    to check_diff_overflow, where the release build was assuming that if
    v1 > v0, v0 - v1 < 0, which is acceptable because v0 - v1 has
    overflown, at which point C says the value is unpredictable.
    
    The fix is to detect the overflow before it happens.

base/gxpflat.c

new errors:

errors that went away:

errors that are not whitelisted:
=== ghostpdl/examples/transparency_example.ps === cp50 ===
=== ghostpdl/examples/transparency_example.ps === psdrgb ===