If one generates an EPS file from the PDF in Acrobat 4, Ghostscript renders this file with the characters shifted up, but the EPS created by Acrobat5 looks fine.
Created attachment 378 [details] acro4_gs811.tar.gz
Wanted to run it with CPSI, but could not unpack the TAR file on Windows. gzip suceeds but tar fails. Could you re-pack it please ?
I've got the files from Alex. CPSI renders both files fine. I confirm the problem with Ghostscript.
GS_7_0X renders fine.
gs8_00 renders it fine. Interesting that in the wrong rendering with CVS HEAD the lower bounary of a character bbox appears equal to the upper boundary of the correct rendering with gs8.00.
Created attachment 385 [details] acrobat4-.eps Attacing a reduced test.
This problem was introduced on 2003-01-27 with activation of the new hinter.
The test document embeds a font, which has a big Y component of lsb. The old code (NEW_TYPE1_HINTER 0) ignores the y-component of lsb with the command ce1_sbw. I have no idea why it was done so. The new code (NEW_TYPE1_HINTER 1) does not ignore it, and we get the shift.
The previous comment isn't correct. The old code accounts the Y coordinate at the end of gs_type1_finish_init. Still didn't find where the new code diverges.
Well, I found the reason. The new hinter imports the side bearing when import_shift is not initialized. Need to initialize it in t1_hinter__init rather than in t1_hinter__set_mapping.
Created attachment 386 [details] patch.txt The attached patch fixes the problem.
An improved patch http://www.ghostscript.com/pipermail/gs-code-review/2003-December/004275.html
Created attachment 388 [details] patch2.txt Attaching an improved patch.
Created attachment 390 [details] patch3.txt Attaching an improved patch3.txt (same as http://www.ghostscript.com/pipermail/gs-code-review/2003-December/004276.html)
Patch http://www.ghostscript.com/pipermail/gs-cvs/2003-December/003799.html has been committed.