news.htm

GhostScript的源代码

halftone parameters.  (gxclrast.c)
	- Banding devices using multi-screen halftones didn't set the
halftone phase correctly.  (gxcht.c, gxclrast.c)
	- RasterOp didn't work for 2- or 4-bit gray-scale devices.  It still
doesn't work right, but it's closer.  (gdevm2.c, gdevm4.c, gdevmrop.c)
	- Undefined characters in CMap-based composite fonts caused errors
and possible invalid memory accesses.  (gschar0.c)

<hr>

<h1><a name="Version5.34"></a>Version 5.34 (beta) (8/16/98)</h1>

More updates for 5.3x.

<h2><a name="V5.34_Documentation"></a>Documentation</h2>

Updates all of the man pages to new versions.  (gs-vms.hlp, gs.1, pdf2dsc.1,
pdf2ps.1, ps2ascii.1, ps2epsi.1, ps2pdf.1, ps2ps.1)

<h2><a name="V5.34_Procedures"></a>Procedures</h2>

Fixes problems:
	- The MSVC library makefile incorrectly included gp_msio.c.
(msvclib.mak, winlib.mak)

<h2><a name="V5.34_Drivers"></a>Drivers</h2>

Fixes problems:
	- The PDF writer unnecessarily checked for other parameters after a
failing show or pdfmark.  (gdevpdfp.c)
	- The device's BufferSpace parameter was ignored.  (bug introduced
by async rendering code in 5.30) (gdevprn.c)
	- Banding devices smashed the first 4 bytes of each scan line.  (bug
introduced in 5.28) (gxdevmem.h, gdevmem.c, gxclread.c)
	- The phase for colored halftones was set wrong.  (We have "fixed"
this innumerable times, and it is still probably wrong.)  (gxclrast.c)

Replaces the deprecated gx_device_image_xxx calls with the new gx_image_xxx
calls.  (gdevpdfi.c, gdevps.c, gdevpx.c, gdevvec.c)

<h2><a name="V5.34_Interpreter"></a>Interpreter</h2>

Fixes problems:
	- The equality test for idiom recognition sometimes incorrectly
returned false.  (zmisc3.c)
	- shfill didn't take the BBox into account.  (It still doesn't.)
(gxshade.c)
	- The status operator could return an incorrect pages or bytes value
for files larger than 2 Gb, rather than giving an error.  (zfile.c)

<h2><a name="V5.34_Library"></a>Library</h2>

Fixes problems:
	- The hint tables weren't reset after processing the base character
of a 'seac', leading to incorrect hinting and (for Type 2 CharStrings)
possible mis-parsing of the outline data.  (gstype2.c)
	- When processing a 'seac', the left side bearing was added in
twice, displacing characters to the right.  (bug probably introduced in
5.20) (gstype1.c)
	- RasterOps didn't set the size of the RasterOp texture device,
which caused filling operations involving a colored halftone to be clipped
to an empty rectangle.  (gdevrops.c)
	- When filling with a colored halftone, the RasterOp source was
often disregarded.  (gxcht.c)
	- gx_device_image[_plane]_data and gx_device_end_image took an
unused dev argument, leading to potential confusion.  We deprecated the
gx_device_image_plane_data and gx_device_end_image procedures, replacing
them with macros that call their replacements gx_image_plane_data and
gx_image_end (which no longer have the spurious dev argument).  (gxdevcli.h,
gxdevice.h, gxiparam.h, gdevbbox.c, gdevddrw.c, gdevdflt.c, gsimage.c,
gslib.c, gxclimag.c, gxclrast.c, gximage2.c, gximage3.c, gximage4.c)
	- gs_color_adjust_count, introduced in 5.33 to address the problem
of freeing free-standing patterns, just made things worse.  We are replacing
them with gs_pattern_reference, essentially the procedure introduced in 5.31
that we deleted in 5.33.  (gsccolor.h, gspcolor.h, gscolor.c, gspcolor.c)

Changes hint suppression from a #define to a debugger-settable static.
(gxhint1.c)

<hr>

<h1><a name="Version5.33"></a>Version 5.33 (beta) (8/14/98)</h1>

This is the third set of updates for 5.30.  These updates have been driven
mostly by the needs of another Aladdin project, so they haven't included as
many problem fixes as we would normally make.

<h2><a name="V5.33_Platforms"></a>Platforms</h2>

Fixes problems:
	- The thread creation procedure for Win32 called a Windows function
rather than the proper C library function.  (gp_win32.c)
	- In some MS-DOS implementations, the copy command doesn't recognize
the /Y switch.  (bcwin32.mak)

<h2><a name="V5.33_Fonts"></a>Fonts</h2>

Fixes problems:
	- TrueType fonts with *less* than 64K of glyf data stopped working.
(bug introduced in 5.31) (gs_ttf.ps)

<h2><a name="V5.33_Interpreter_PDF"></a>Interpreter (PDF)</h2>

Fixes problems:
	- Using an Identity function left an extra element on the stack.
(pdf_draw.ps)
	- The syntax of # in names is different in PDF 1.1 and PDF 1.2, but
the PDF 1.2 syntax was always used.  (pdf_base.ps, pdf_main.ps)

<h2><a name="V5.33_Library"></a>Library</h2>

Fixes problems:
	- Color images still always caused an error. (bug introduced in
5.32) (gxicolor.c)
	- Because of a typo, the implementation of image types didn't
compile. (bug introduced in 5.32)  (gxiparam.h)
	- Free-standing colors (specifically pattern colors) and color
spaces couldn't be freed.  This required a NON-BACKWARD-COMPATIBLE change in
an internal API used for managing color and color space reference counts.
(gxcspace.h, gsccolor.h, gscdevn.c, gscolor.c, gscolor2.c, gscscie.c,
gscsepr.c, gscspace.c, gspcolor.c)

Cleans up a little more formatting.  (gscspace.h)

<hr>

<h1><a name="Version5.32"></a>Version 5.32 (beta) (8/13/98)</h1>

This is the second set of updates for 5.30.

<h2><a name="V5.32_Procedures"></a>Procedures</h2>

Fixes problems:
	- The OS/2 makefile accidentally omitted the OS/2-PM driver.
(os2.mak)
	- The MSVC library makefile didn't work.  (msvclib.mak)
	- Debugging was turned on in the Borland Windows makefile.
(bcwin32.mak)
	- The BEGINFILES2 list was too long in the MSVC makefile.
(msvc32.mak)

<h2><a name="V5.32_Interpreter"></a>Interpreter</h2>

Fixes problems:
	- .execn wouldn't accept pseudo-operators.  (iref.h)
	- inustroke with a matrix operand didn't work correctly.  (zupath.c)
	- ImageType 4 images didn't work.  (bug introduced in 5.31)
(zimage3.c)

<h2><a name="V5.32_Interpreter_PDF"></a>Interpreter (PDF)</h2>

Fixes problems:
	- Forms used more than once caused an invalidaccess error.
(pdf_draw.ps)

<h2><a name="V5.32_Library"></a>Library</h2>

Fixes problems:
	- Using a colored pattern with a Pattern color space that included a
base space could access uninitialized floating point values, possibly
causing an arithmetic trap.  (gspcolor.c)
	- RasterOp wasn't implemented for 16- and 32- bit devices.
(gdevm16.c, gdevm32.c, gdevmrop.c)
	- The library test program relied on copysign, which some C
libraries lack.  (gslib.c)
	- The gs_free and gs_heap_free_all debugging printout were included
even if DEBUG wasn't defined. (bug introduced in 5.30) (gsmalloc.c)
	- Color images always caused an error. (bug introduced in 5.30)
(gxicolor.c)
	- Masked images with pis = NULL could cause an error.  (gximono.c)
	- In landscape orientation, masked images didn't guarantee that the
buffer of mask scan lines would be flushed to the mask clipping device
before processing the data being masked.  (gxiparam.h, gxidata.c, gxifast.c,
gximage3.c)
	- Images with source transparency set but CombineWithColor = false
could get an error because of trying to paint with an unloaded
color. (caused by a bug fix in 5.30) (gxdcolor.c, gxiinit.c)

<hr>

<h1><a name="Version5.31"></a>Version 5.31 (limited) (8/11/98)</h1>

This is the first set of updates for 5.30.  They fix a few minor build
problems, and some significant problems with bitmap patterns that do not
affect the PostScript or PDF interpreters.

Missing DPS functionality:
	Contexts:
		Free L/G VM when num_contexts goes to 0
		quit should terminate only the current context
		  Contexts terminated on quit or error destroy stacks &
		    close stdin/out
		Monitor-lock oparrays?
		Global VM GC should collect (or at least scan) *all* local
		  VMs that reference it
		How to handle multi-client input?
	Type 2 images:
		UnpaintedPath (generating X exposure events)
		Transformed source
		Overlapping source/dest (affects order)
	NeXT extensions:
		Compositing
		    Storing alpha with X windows
		    Halftoning

Other partially implemented functionality:
  Internal:
	Directory restructuring
	  Create .dev files in GEN directories
	  'make clean' do the right thing
	Typed images in band list
	  (wrong in gxclimag.c, not implemented in gxclread.c)
	General image_plane_data for banding
	alpha/compositing-related:
	  Image alpha data in band list (gxclimag.c?)
	  Compositing through band list (stubbed in gxclimag.c)
	Driver text API (barely started, not documented)
  PDF-related:
	Acrobat Distiller 3.0 pdfmarks (see under Drivers) - untested
  PostScript Level 2+ leftovers:
	HalftoneTypes 6, 10, and 16 (in gs_ll3.ps)
	Type 32 fonts (gs_typ32.ps, zchar32.c, zfont32.c)
  PostScript LanguageLevel 3:
	clipsave/cliprestore (dummy)
	DeviceN color space (zcsdevn.c, gscdevn.c)
	  How to handle arbitrary numbers of colorants?
	  Caching transformed values
	ImageType 3 images (zimage3.c)
	  Library doesn't check ImageMatrix compatibility
	    (see Table 4.24 in PS3 RM draft)
	DevicePixel color space
	  Not enough bits in client or concrete color!
	  What is initial value?
	PatternTypes other than 1
	Functions
	  Monotonicity is implemented only for Exponential and ArrayedOutput
	    (may not be needed at all....)
	Shading
	  Doesn't handle Background, BBox, Extend, or clipping
	Streams
	  LZWDecode LowBitFirst
	  ReusableStreamDecode
	UseCIEColor (see Section 4.8.3 in PS3 RM draft)

Known problems:
	- Setting HWSize by itself with setpagedevice has no effect.
	- When shrinking oversize halftone cells, the current algorithm,
which progressively halves the size, can produce non-monotonic behavior; GS
should either use a monotonic algorithm or give a limitcheck.
(k/jdscreen.ps)
	- In Type 1 fonts, almost-vertical lines that straddle a stem edge
produce uneven output.
	- At some resolutions, gaps occur in some character outlines with
certain Type 1 fonts.  (? - no test file)
	- The Watcom MS Windows makefile (watcw32.mak) doesn't work.
	- The uniprint driver sometimes drops some output, and is very slow
at 720x720 resolution.
	- Halftones look relatively coarse: we should be using a small-scale
version of AccurateScreens (multiple small dot cells with an ordered dither
cell visiting pattern) to get both fine screens and many levels.
	- Since the x11alpha device uses saturation to emulate alpha,
writing anti-aliased objects over dark areas produces incorrect output.
(gdevxalt.c)
	- Setting OutputFile=- should redirect stdout to stderr (or provide
some other way to redirect stdout).
	- There is a problem on page 3 of k/microw-j.pdf, related to Type 0
fonts.
	- Ghostscript can't handle double-byte TrueType fonts.
(/import/fonts/cyberbit/CYBERBIT.TTF)
	- High-level output devices ([e]pswrite, pdfwrite, pxl*) always
produce a single output file rather than a separate file for each page, even
if there is a %d in the OutputFile name.
	- ps2pdf doesn't handle the Distiller 3.0 BP/EP/SP pdfmarks.
(k/pdfmark3.ps)
	- CTM is not set correctly for PostScript pass-throughs in the PDF
interpreter.
	- When attempting both RasterOp and anti-aliasing, the anti-aliasing
takes effect rather than the RasterOp.
	- Selecting the nullpage device initially, gsave, change page
device, grestore produces infinite recursion.
	- The MSVC library and Watcom makefiles are broken.
	- copy_rop doesn't work for 2- or 4-bit devices, because monobit
copy_rop may call other device procedures (fill_rectangle, copy_mono,
tile_rectangle).

<h2><a name="V5.31_Drivers"></a>Drivers</h2>

Moves some casting macros from generic files to the few specific files that
use them.  (gdev4693.c, gdevn533.c, gdevsnfb.c)

<h2><a name="V5.31_Interpreter"></a>Interpreter</h2>

Changes ImageType 4 images to match the change in the library interface.
(zimage3.c)

<h2><a name="V5.31_Library"></a>Library</h2>

Fixes bugs:
	- The MaskColor elements of ImageType 4 images were defined as int
rather than uint, truncating the range by 1 bit and causing problems in
comparing the mask color against the pixel samples.  (gsiparm4.h,
gximage4.c)
	- A needed cast was missing from gs_malloc_memory_release.
(gsmalloc.h)
	- A needed cast from byte to char was missing.  (gsparams.c)
	- Patterns always allocated a mask for rendering, which for large
patterns specified directly as bitmaps allocated a lot of unnecessary
memory.  (gxcolor2.h, gspcolor.c, gxp1fill.c, gxpcmap.c)
	- Macros simply defining casted versions of variables made the code
hard to debug.  We have only removed the ones that are in generic code (not
drivers) and that cast pointers to gx_device or its subclasses: the only
ones removed from header files were prn_dev in gdevprn.h, mdev in gdevmem.h,
and ccdev, crdev, and cwdev in gxcldev.h.  gdevmem.h, (gdevprn.h, gxclist.h,
gdevabuf.c, gdevm1.c, gdevm2.c, gdevm4.c, gdevm8.c, gdevm16.c, gdevm24.c,
gdevm32.c, gdevmpla.c, gdevnfwd.c, gdevprn.c, gxacpath.c, gxclimag.c,
gxclist.c, gxclpage.c, gxclpath.c, gxclread.c, gxclrect.c)
	- There was no way to free a pattern instance outside the graphics
state.  (gspcolor.h, gspcolor.c)

<hr>

<h1><a name="Version5.30"></a>Version 5.30 (limited) (8/10/98)</h1>

This is the first beta candidate for the 5.50 release.  It completes the
merge of the overlapped rasterizing code into the main development line.