In Comment #5 for Bug 692770 <http://bugs.ghostscript.com/show_bug.cgi?id=692770#c5> I mentioned that Ghostscript compiled with clang generates different output for a file than Ghostscript compiled with gcc. I've now gone ahead and compared clang vs. gcc compiled Ghostscript for the entire regression test suite and found differences in the following files when converted directly to bitmaps: tests_private__comparefiles__046-01.ps (page 1) tests_private__comparefiles__276-01.ps (page 1) tests_private__comparefiles__Bug689006.pdf (page 1) tests_private__comparefiles__Bug690179.pdf (page 1) tests_private__ps__ps3cet__09-30.PS (page 5) tests_private__ps__ps3cet__09-31.PS (page 10) tests_private__ps__ps3cet__09-41.PS (page 4) tests_private__ps__ps3cet__09-48.PS (page 4) and in these files when converted to PDF via pdfwrite and then to bitmap: tests_private__comparefiles__Bug692634.ps (page 1) tests_private__ps__ps3cet__18-02G.PS (page 3) I've attached a .tar file with scaled bitmaps of the files with differences. The left third is the file processed by gcc compiled Ghostscript, the middle third is the file processed by clang compiled Ghostscript, and the right third is the difference.
In addition to differences in output with the clang compiled gs vs the gcc compiled gcc, these test files seg fault with the clang compiled gcc: tests_private__comparefiles__Bug692634.ps tests_private__ps__ps3cet__09-02.PS tests_private__ps__ps3cet__09-28.PS tests_private__ps__ps3cet__09-29.PS tests_private__ps__ps3cet__09-52.PS
With the current code, 5596cc435aec8387ddd2e64437f1f9486f3ee5c6, the differences from comment #0 and the segfaults from comment #2 no longer occur.