Hi

I tested the newest version (pornel-improved-pngquant-1.8.3-108-gf309d1b) to compress drawings and got a bad result.
Test on:
openSUSE 12.1 (x86_64) and on SUSE LINUX 10.1 (i586)

without any further options:
pngquant 20094695E001101.a.png

Best regards
Horst

1. Unfinished code in libimagequant.c
   LIQ_EXPORT liq_attr* liq_attr_create_with_allocator(void* (_malloc)(size_t), void (_free)(void*))
   {
   ...
   if ((malloc != NULL) != (free != NULL)) return NULL; // either specify both or none

liq_attr *attr = malloc(sizeof(liq_attr));

...
If both malloc and free are NULL a problem results (attempt to use NULL function pointer) .
The comment suggests there were plans to handle that case, but they were not implemented.
Not a real problem, just a question mark in code analyser.

2. Memory leak in pngquant.c
   outname is never free()d.
   Since outname is to be free()d just before the program exit that is not a real problem,
   but since other memory areas are properly deallocated, adding free(outname) just befor "return" from "main" would be nice.

3. pngquant is not friendly to misuse as a library
   I misused pngquant as a library, renaming its "main" routine and calling it from my code with suitable arguments in argv. The second call fails due to getopt being in need of a reset.
   Consider adding reset of getopt
   optind = 1;
   optreset = 1;
   at the start of the main routine.
   It will not affect the intended use (one run as a standalone program), but will help in my scenario.

Sincerely, Michail

commit 35d8cda breaks compilation:
rwpng.c: In function 'rwpng_write_image24':
rwpng.c:323: error: 'png_image' undeclared (first use in this function)

https://github.com/pornel/improved-pngquant/blob/master/rwpng.c#L323
rename png_image -> rwpng_png_image

pngquant --force --ext=.PNG -- file.PNG

creates a new file called file.PNG.PNG

pngquant --force --ext=.png -- file.png

overwrites file.png

Would it be possible to get --ext=.PNG to overwrite too?

pngquant-improved, version 1.4.3.1b (August 2011), by Greg Roelofs, Kornel Lesinski.
Compiled with libpng 1.2.37; using libpng 1.2.37.
Compiled with zlib 1.2.3; using zlib 1.2.3.

Windows XP SP3, Windows 7 SP1

pngquanti.exe -verbose 256 img.png
img.png:
Reading file corrected for gamma 2.2
making histogram...

http://sklabs.ru/tmp/img.png

I am using ImageAlpha but now i want to compress all of the files in batch but ImageAlpha compress more then using terminal command "pngquant 256 *.png" and cant find the other argument that need to send with this like "Dithered" and "Median Cut".
I am using Image Alpha with default settings. please let me know about more arguments.

Hi,

I'm trying to use "improved-pngquant" in my build script. I was looking for a way to overwrite old file with a new (compressed) file.

I found out thru "-h" option that there is a -f option but that does not have any effect and I'm still getting default extension "-fs8".

Is there any work around on that?

Hi,

I am trying to compile your latest version on Centos and am unable to.

[root@sydney pornel-improved-pngquant-0af1c98]# make
gcc -DNDEBUG -Wall -Wno-unknown-pragmas -I. -I../libpng -I../zlib -I/usr/local/include/ -I/usr/include/ -I/usr/X11/include/ -O3 -mtune=native -fearly-inlining -fstrict-aliasing -ffast-math -funroll-loops -fstrict-overflow -fomit-frame-pointer -momit-leaf-frame-pointer -ffinite-math-only -fno-trapping-math -funsafe-loop-optimizations -std=c99 -c -o pngquant.o pngquant.c
cc1: error: unrecognized command line option "-fstrict-overflow"
pngquant.c:1: error: bad value (native) for -mtune= switch
make: *** [pngquant.o] Error 1

I've removed both -mtune and -fstrict-overflow from the makefile, but then when I try to compile, I get a number of errors. The initial one is:

[root@sydney pornel-improved-pngquant-0af1c98]# make
gcc -DNDEBUG -Wall -Wno-unknown-pragmas -I. -I../libpng -I../zlib -I/usr/local/include/ -I/usr/include/ -I/usr/X11/include/ -O3 -fearly-inlining -fstrict-aliasing -ffast-math -funroll-loops -fomit-frame-pointer -momit-leaf-frame-pointer -ffinite-math-only -fno-trapping-math -funsafe-loop-optimizations -std=c99 -c -o pngquant.o pngquant.c
In file included from pngquant.c:56:
rwpng.h:35:57: error: png.h: No such file or directory
In file included from pngquant.c:56:
rwpng.h:60: error: expected specifier-qualifier-list before ‘png_uint_32’
rwpng.h:70: error: expected specifier-qualifier-list before ‘png_uint_32’

I don't know if png.h is supposed to be included in the distribution or it should already be on the system. Can you please let me know what the dependencies are?

I think this is a brilliant program and I'd like to use it.

Thanks,
Tom

Hello,

I have just compiled 1.8.3 (with libpng 1.5.14) and filtered output looks like not filtered output. Results from 1.8.2 version were much better.
See following example, 16 colors.
1.8.2 filtered

1.8.3 filtered

1.8.2 not filtered

Have I missed something ?

Have a nice day,

    AiZ

thanks for an awesome tool. it would be great if it could recognize 9-patched images and preserve the 9-patch formatting.
imo this functionality would make this the tool of choice for android development. here is some info about 9pathes if you're not familiar:
http://developer.android.com/guide/topics/graphics/2d-graphics.html#nine-patch
http://developer.android.com/tools/help/draw9patch.html

pngquant, version 1.6.3 (December 2011), by Greg Roelofs, Kornel Lesinski.
Compiled with libpng 1.2.44; using libpng 1.2.44.
Compiled with zlib 1.2.3.4; using zlib 1.2.3.4.

uname -a:
Linux HOST 2.6.32-5-xen-amd64 #1 SMP Tue Mar 8 00:01:30 UTC 2011 x86_64 GNU/Linux

Program received signal SIGSEGV, Segmentation fault.
0x0000000000406d69 in pam_computeacolorhash (apixels=0x611850, cols=290,
rows=2, gamma=0.45455000000000001, maxacolors=1966080, ignorebits=0,
importance_map=0x611874, acolorsP=0x7fffffffe38c) at pam.c:80
80 union rgb_as_long px = {apixels[row][col]};
(gdb) bt
#0 0x0000000000406d69 in pam_computeacolorhash (apixels=0x611850, cols=290,

rows=2, gamma=0.45455000000000001, maxacolors=1966080, ignorebits=0, 
importance_map=0x611874, acolorsP=0x7fffffffe38c) at pam.c:80

#1 0x0000000000406c41 in pam_computeacolorhist (apixels=0x611850, cols=290,

rows=2, gamma=0.45455000000000001, maxacolors=1966080, ignorebits=0, 
importance_map=0x611870) at pam.c:52

#2 0x00000000004046e1 in histogram (input_image=0x7fffffffea00,

reqcolors=256, speed_tradeoff=3, importance_map=0x611870) at pngquant.c:634

#3 0x0000000000405bc5 in pngquant (input_image=0x7fffffffea00,

output_image=0x7fffffffe4f0, options=0x7fffffffeb00) at pngquant.c:931

#4 0x0000000000402915 in main (argc=3, argv=0x7fffffffec38) at pngquant.c:223

(gdb)

Regards,
Oleg

We are using 1.8.1 on some png and it is doing an amazing job (thank you for the great work), particularly in alpha gradient areas. However, I just downloaded 1.8.3 on my mac, and the results it produced were noticeably worse than that of 1.8.1, and slightly larger file size, at least for the example below.
This is a sample result with 1.8.1

and this is the result with 1.8.3

you can see banding in the shadows (which use alpha) in the 1.8.3 output

This is the command line I used:
./pngquant 256 a-christmas-carol-2013-02-27-123019.png

ps- I know, Marley's ghost probably should not be casting a shadow ;-)

I know that this kind of bugs sucks, but I'm stuck. Errors:

pornel-improved-pngquant-93d8bff]$ make
gcc -Wall -Wno-unknown-pragmas -I. -I../libpng -I/usr/local/include/ -I/usr/include/ -I/usr/X11/include/ -DNDEBUG -O3 -fstrict-aliasing -ffast-math -funroll-loops -fomit-frame-pointer -ffinite-math-only -std=c99 -c -o pngquant.o pngquant.c
gcc -Wall -Wno-unknown-pragmas -I. -I../libpng -I/usr/local/include/ -I/usr/include/ -I/usr/X11/include/ -DNDEBUG -O3 -fstrict-aliasing -ffast-math -funroll-loops -fomit-frame-pointer -ffinite-math-only -std=c99 -c -o rwpng.o rwpng.c
gcc -Wall -Wno-unknown-pragmas -I. -I../libpng -I/usr/local/include/ -I/usr/include/ -I/usr/X11/include/ -DNDEBUG -O3 -fstrict-aliasing -ffast-math -funroll-loops -fomit-frame-pointer -ffinite-math-only -std=c99 -c -o pam.o pam.c
gcc -Wall -Wno-unknown-pragmas -I. -I../libpng -I/usr/local/include/ -I/usr/include/ -I/usr/X11/include/ -DNDEBUG -O3 -fstrict-aliasing -ffast-math -funroll-loops -fomit-frame-pointer -ffinite-math-only -std=c99 -c -o mediancut.o mediancut.c
pam.h: In function `colordifference_ch':
mediancut.c:439: error: prior parameter's size depends on 'clrs'
make: *** [mediancut.o] Error 1

I have zlib-devel and libpng-devel installed

$ rpm -qa | grep -e 'zlib' -e 'libpng'
zlib-1.2.1.2-1
zlib-devel-1.2.1.2-1
libpng-1.2.7-1
libpng-devel-1.2.7-1

Full RHEL version

$ cat /etc/redhat-release
Red Hat Enterprise Linux ES release 4 (Nahant Update 1)

Thanks in advance for your help!

Original downstream report is here:

https://bugs.gentoo.org/show_bug.cgi?id=464762

Complete build.log is here:

https://bugs.gentoo.org/attachment.cgi?id=344534

And the patch that fixes the conflict is here:

http://sources.gentoo.org/cgi-bin/viewvc.cgi/gentoo-x86/media-gfx/pngquant/files/pngquant-1.8.3-libpng16.patch?view=markup

The reason is obvious, libpng 1.6 now has it's own png_image and it's conflicting with the png_image that comes with the rwpng.{h,c}
I expect any software making use of these 2 source files to fail

Reason is in line 371 ../libpng/pngconf.h, solution:

diff --git a/pngquant.c b/pngquant.c
index de1bd81..f895859 100644
--- a/pngquant.c
+++ b/pngquant.c
@@ -52,6 +52,7 @@

include <io.h> /* setmode() */

#endif

+#define PNG_SKIP_SETJMP_CHECK
#include "rwpng.h" /* typedefs, common macros, public prototypes */
#include "pam.h"
#include "mediancut.h"

Tested successfully on Windows 7 64 bits, compiled with libpng 1.6.3 and zlib 1.2.8.
Link: http://css-ig.net/tools/zip/pngquantv2-1.6.3.zip

When I optimized PNGs on Mavericks it really desaturates the colors. I tested this across many different PNGs and it does it to all of them.

i tried to compile on Debian Lenny which did not work properly:
gcc -std=c99 -O3 -Wall -I. -I../libpng -I../zlib -I/usr/X11/include/ -funroll-loops -fomit-frame-pointer -c -o pam.o pam.c
pam.c: In function ‘pam_hashapixel’:
pam.c:107: error: ‘register_t’ undeclared (first use in this function)
pam.c:107: error: (Each undeclared identifier is reported only once
pam.c:107: error: for each function it appears in.)
pam.c:112: error: expected specifier-qualifier-list before ‘register_t’
pam.c:117: error: ‘struct ’ has no member named ‘l1’
pam.c:117: error: ‘struct ’ has no member named ‘l1’
pam.c:117: error: ‘struct ’ has no member named ‘l2’
make: *** [pam.o] Error 1

if it remove the "-std=c99" it compiles without problems.

After that i got an error while linking though:
gcc -o pngquant pngquant.o rwpng.o pam.o -L../libpng -L../zlib -L/usr/X11/lib/ -L/usr/lib/ -lz -lpng -lm
pngquant.o: In function pngquant': pngquant.c:(.text+0x3a15): undefined reference tomergesort'
pngquant.c:(.text+0x44d0): undefined reference to `mergesort'
collect2: ld returned 1 exit status

i "fixed" this by reverting 911e35b

After this change it compiles fine.

i currently have have:
ii zlib1g 1:1.2.3.3.dfsg-12 compression library - runtime
ii zlib1g-dev 1:1.2.3.3.dfsg-12 compression library - development
ii libpng12-0 1.2.27-2+lenny4 PNG library - runtime
ii libpng12-dev 1.2.27-2+lenny4 PNG library - development
ii gcc-4.3-base 4.3.2-1.1 The GNU Compiler Collection (base package)
installed

i tried to compile the same version on Mac OSX Snow Leopard which works fine without any changes, but there gcc 4.2.1 is used

Using pngquant with quality 100 on the same file, colors can get reduced for a few iterations.

$ pngquant -f --speed 1 --ext .png -v --quality 100- file.png
  made histogram...16 colors found
  mapped image to new colors...MSE=0.002 (Q=100)
  writing 15-color image as file.png

$ pngquant -f --speed 1 --ext .png -v --quality 100- file.png
  made histogram...15 colors found
  mapped image to new colors...MSE=0.022 (Q=100)
  writing 14-color image as file.png

$ pngquant -f --speed 1 --ext .png -v --quality 100- file.png
  made histogram...14 colors found
  mapped image to new colors...MSE=0.000 (Q=100)
  writing 14-color image as file.png

This produces the exact same file, but at lower reported quality.

$ pngquant 14 -f --speed 1 --ext .png -v file.png
  made histogram...16 colors found
  mapped image to new colors...MSE=0.076 (Q=99)
  writing 14-color image as file.png

Hi,

Firstly, thumbs-up for this util!

I have a problem building from source when using USE_COCOA=1

fatal error: 'CoreGraphics/CoreGraphics.h' file not found

(in v2 there is an extra error in the Makefile)

The solution is to change the import in rwpng_cocoa.m:

-#import <CoreGraphics/CoreGraphics.h>
+#include <ApplicationServices/ApplicationServices.h>

Want a pull request to fix this issue?

p.s. why the colors appear a bit desaturated when using the cocoa build? Something to do with the color profiles? Check here http://dabblet.com/gist/6351700

Hello!

When trying the latest version of pngquant (from master), I noticed that in some cases an even colour is being dithered. This does not happen in 1.8.3. It has an extremely negative effect on the image size. In this specific example the image size of the 2.0-optimised version is almost 3 times larger than the 1.8.3-optimised version!

Here is an example of the dithering in 2.0:

Here is the result from 1.8.3 (51 KB).

Here is the result from 2.0 (138 KB).

Here is the original (187 KB).

The images were produced with pngquant 256 example.png.

Do you have any idea what the cause is? Is this intended behaviour with 2.0, or is it simply not ready yet?

I tried feeding a 1-bit (monochrome) PNG through pngquant. The output was converted to 8-bit and the image was completely corrupt. I can send example files if required.

I assume that pngquant should not increase the palette size of the input PNG.

Hi, I have noticed that a .png image I am tying to shrink, is maintaining the feathering very well except where color is feathered into transparent in which case the pixels are either coming out fully opaque or fully transparent.

E.g...
https://skydrive.live.com/redir?resid=9F8DE44BC9AC7FD6!120&authkey=!AHEnMrghCFH2YEU

after running: pngquant 256 slide-jigsaw.png

https://skydrive.live.com/redir?resid=9F8DE44BC9AC7FD6!119&authkey=!APCohP1u4LKAYtQ

version 1.7.2
I'm on windows 7 64-bit

Fancy random() can be replaced with more portable rand().

Here is patch for 1.7.2 to use standard option syntax. For in-depth details, see: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=602585

 git remote add jaalto git://github.com/jaalto/improved-pngquant.git
 git fetch jaalto
 git diff jaalto/options

 git merge jaalto/options

Hello, I compiled the library with GCC from Cygwin on Windows, and add libimagequant.a along with libgcc.a to my VC project, use:
#pragma comment(lib, "./lib/libgcc.a")
#pragma comment(lib, "./lib/libimagequant.a")

liq_attr *attr = liq_attr_create();

At the end of the link I get the following errors:

Linking...
libimagequant.a(libimagequant.o) : error LNK2001: unresolved external symbol __f_pow
libimagequant.a(pam.o) : error LNK2001: unresolved external symbol __f_pow
libimagequant.a(libimagequant.o) : error LNK2001: unresolved external symbol __f_powf
libimagequant.a(mediancut.o) : error LNK2001: unresolved external symbol __f_powf
libimagequant.a(libimagequant.o) : error LNK2001: unresolved external symbol ___getreent
libimagequant.a(libimagequant.o) : error LNK2001: unresolved external symbol _vsnprintf
Debug/RenderSizeTest.exe : fatal error LNK1120: 4 unresolved externals
Error executing link.exe.

how can i fix the issue? thanks!

So I have a program that uses pngquant to process from one directory and output the compressed files to another directory. The new --output flag allowed me to not having to make a copy of the source directory. That's great. But I realize if the destination folder and its subdirectories do not exist before I call pngquant, I'd run into the not-writeable error in write_image. I wonder if there is an easy way to make it work even if the destination isn't there (maybe only do this when it's on both force&&output mode)

Hi porneL - wouldn't compile on MinGW on Windows, so I created a fork that fixes the issues I ran into (including detailed INSTALL instructions). Hope this is useful to other Windows users, as there's currently no Windows binaries online!

https://github.com/neogic/improved-pngquant-mingw

Cheers, Ben
Neogic Web Solutions
http://www.neogic.com

The homepage states that the current version of pngquant is 2.0.1, yet there is no tarball or git tag for that version.

hi to all
i want to download script tiny png for website
please help me

very thank

Hello,

One Grunt plugin is using your tool and on linux-x86-64 it is throwing this error:

/path/to/node_modules/grunt-contrib-imagemin/node_modules/pngquant-bin/vendor/pngquant: /lib64/libc.so.6: version `GLIBC_2.11' not found

Linux is a CI machine. I contacted the admin and here is what he replied:
I ran updates to glibc from our Enterprise Linux distribution but it did not help with this problem. :(

First I wanted to post this as a Grunt plugin issue, but I guess it is more related to the pngquant itself.
Can you give some tips on resolving this?

Hello,

First of all, I would like to say that this program is the best png optimizer I have ever found. Thanks a lot for it.

I was wondering if it would be possible for you to create a build for Linux C CGI ?
That would be useful to automatically reduce the size of png files users upload on their website.

I have some advanced knowledge in C, but I can't figure out how the command line flags are working in your program...

Thank you for your time.

With some images, pngquant appears to poorly align the colormap.

Before:
After:

They look similar, but the former has a transparent background, and the latter has a white background. This is a major issue for my application, which layers images on top of others.

System (CentOS5.2):

Linux 2.6.21.7-2.fc8xen #1 SMP Fri Feb 15 12:34:28 EST 2008 x86_64 x86_64 x86_64 GNU/Linux

These are the libpng-devel versions installed:

libpng-devel.i386 2:1.2.10-7.1.el5_0.1 libpng-devel.x86_64 2:1.2.10-7.1.el5_0.1

Make gives us this output:

gcc -DNDEBUG -O3 -Wall -I. -I../libpng -I../zlib -I/usr/X11/include/ -funroll-loops -fomit-frame-pointer -std=gnu99   -c -o pngquant.o pngquant.c
In file included from /usr/include/png.h:371,
                 from rwpng.h:34,
                 from pngquant.c:46:
/usr/include/pngconf.h:336: error: expected '=', ',', ';', 'asm' or '__attribute__' before '.' token
/usr/include/pngconf.h:337: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'include'
In file included from pngquant.c:49:
nearest.h:8:44: warning: no newline at end of file
pngquant.c: In function 'main':
pngquant.c:222: warning: missing braces around initializer
pngquant.c:222: warning: (near initialization for 'input_image.jmpbuf[0]')
pngquant.c:223: warning: missing braces around initializer
pngquant.c:223: warning: (near initialization for 'output_image.jmpbuf[0]')
make: *** [pngquant.o] Error 1

Seems to be a known problem when compiling libpng with another thing that uses "setjmp.h", the common practice seems to be to patch the pngconf.h under /usr/include... but uh, dunno if I want to do this.
INSTALL of libpng talks about some ways to work around this in code, but I don't have the time to dig into the whole thing right now.

Would be nice if you could comment on this and whether you're planning to "somehow" fix it

Best regards,

Ivo

It should run on Windows 32 or 64 bits (tested on Seven 64 bits)
Download link: http://css-ig.net/tools/zip/pngquant.zip

Maybe this one is better > http://pngquant.org/pngquant-windows.zip

I posterize.c:369:5: warning: implicit declaration of function 'getopt' [-Wimplicit-function-declaration]
posterize.c:374:48: error: 'optarg' undeclared (first use in this function)
posterize.c:374:48: note: each undeclared identifier is reported only once for each function it appears in
posterize.c:382:16: error: 'optind' undeclared (first use in this function)

I add #include <getopt.h> in posterize.c to fix this problem.

/usr/bin/ld: posterize.o: undefined reference to symbol 'floor@@GLIBC_2.2.5'
/usr/bin/ld: note: 'floor@@GLIBC_2.2.5' is defined in DSO /lib64/libm.so.6 so try adding it to the linker command line
/lib64/libm.so.6: could not read symbols: Invalid operation
collect2: error: ld returned 1 exit status
make: *** [posterize] Error 1

I add -lm in Makefile to fix this problem.

My compile env is:
Linux localhost.localdomain 3.8.11-200.fc18.x86_64 #1 SMP Wed May 1 19:44:27 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux

When using grunt pngmin we get two different output png files under Win & Mac OS.

It turned out to be caused by pngquant under different OS.

Check the issue here.
zauni/pngmin#6 (comment)

Hi. I'm working on a project where we export (and re-export) lots of images from photoshop (using Slicy). We always want to run these exported pngs through pngquant at some point, but we usually end up doing this in batches, to all pngs in our assets-directory.

Do the images degrade in quality after running pngquant on the same png multiple times? Our build-system relies on checking the md5-hash of all our dependencies (including assets) to repackage / create a new version if files have changed. Is there any way we could run pngquant on all images in a directory, but only process the ones that have not already been compressed using pngquant? By checking if the image has already been paletted / reduced to 8-bit (or some other way)?

Hi!

I came across a situation where there can be an array index out of bounds in nearest_search(). I'd love to send a pull request but I'm not entirely sure what the problem is.

Here is a link to the test image that causes the observed behaviour: http://www.schaik.com/pngsuite/tbwn3p08.png

This happens when running with: pngquant --quality 0-99 256 tbwn3p08.png

And here is some debug information produced with Valgrind:

Invalid read of size 8
   at 0x10000A3A0: nearest_search (nearest.c:228)
   by 0x1000082C9: viter_do_iteration (viter.c:76)
   by 0x100012463: pngquant_quantize (libimagequant.c:1320)
   by 0x10000D45F: liq_quantize_image (libimagequant.c:593)
 Address 0x1000e3f90 is 0 bytes after a block of size 4,576 alloc'd
   at 0x58DB: calloc (in /usr/local/Cellar/valgrind/3.8.1/lib/valgrind/vgpreload_memcheck-amd64-darwin.so)
   by 0x1000036E0: pam_colormap (pam.c:238)
   by 0x10000653F: colormap_from_boxes (mediancut.c:417)
   by 0x100006493: mediancut (mediancut.c:400)
   by 0x100011DA9: find_best_palette (libimagequant.c:1239)
   by 0x100012330: pngquant_quantize (libimagequant.c:1305)
   by 0x10000D45F: liq_quantize_image (libimagequant.c:593)

Invalid read of size 8
   at 0x10000A3A4: nearest_search (nearest.c:228)
   by 0x1000082C9: viter_do_iteration (viter.c:76)
   by 0x100012463: pngquant_quantize (libimagequant.c:1320)
   by 0x10000D45F: liq_quantize_image (libimagequant.c:593)
 Address 0x1000e3f98 is 8 bytes after a block of size 4,576 alloc'd
   at 0x58DB: calloc (in /usr/local/Cellar/valgrind/3.8.1/lib/valgrind/vgpreload_memcheck-amd64-darwin.so)
   by 0x1000036E0: pam_colormap (pam.c:238)
   by 0x10000653F: colormap_from_boxes (mediancut.c:417)
   by 0x100006493: mediancut (mediancut.c:400)
   by 0x100011DA9: find_best_palette (libimagequant.c:1239)
   by 0x100012330: pngquant_quantize (libimagequant.c:1305)
   by 0x10000D45F: liq_quantize_image (libimagequant.c:593)

Conditional jump or move depends on uninitialised value(s)
   at 0x10000A668: nearest_search (nearest.c:229)
   by 0x1000082C9: viter_do_iteration (viter.c:76)
   by 0x100012463: pngquant_quantize (libimagequant.c:1320)
   by 0x10000D45F: liq_quantize_image (libimagequant.c:593)
 Uninitialised value was created by a heap allocation
   at 0x5237: malloc (in /usr/local/Cellar/valgrind/3.8.1/lib/valgrind/vgpreload_memcheck-amd64-darwin.so)
   by 0x1000079CE: mempool_create (mempool.c:29)
   by 0x100009C6C: nearest_init (nearest.c:135)
   by 0x100008231: viter_do_iteration (viter.c:67)
   by 0x100012463: pngquant_quantize (libimagequant.c:1320)
   by 0x10000D45F: liq_quantize_image (libimagequant.c:593)

I've narrowed it down to the fact that eventually in nearest_search() sometimes palette_index_guess is equal to centroids->map->colors, which means the array read is out of bounds. I don't know why this would happen or where to take it from here, though.

Suggestions?

Discovered pngquant on onethingwell.org. Nice work! I would like to submit an
OpenBSD port, but I'd like to see if we can smooth out the build a bit first:

1. BSD's install doesn't include a -D flag, I think the best thing to do is
   to use @mkdir -p
2. The install section wasn't installing the man page. It may not be possible to
   write this in a portable way since there is no standard location to install
   man pages.
3. MAKEFLAGS may be multiple variables, but -$(MAKEFLAGS) seems to expect one parameter
4. -I. is superfluous
5. Leave in -I/usr/include/local even if the CFLAGS is defined. Is -I/usr/include ever required?
6. I think any ports system will want to define such paramets as -Wall and -Wno-unknown-pragmas

All of these are just suggestions, I've posted my current patch for you to
review at
Link to previous patch
https://github.com/jasperla/openbsd-wip/blob/adcbd312f10b5c65772ecf66ad65bb843056e9b6/graphics/pngquant/patches/patch-Makefile

There is no recent Ubuntu package, so I tried to build from source. I got the following error after running "make install"

In file included from pngquant.c:56:0:
rwpng.h:35:57: fatal error: png.h: No such file or directory
compilation terminated.
make: *** [pngquant.o] Error 1

I had a quick look a rwpng.h and it contains

#include "png.h"    /* libpng header; includes zlib.h */

But there is no png.h in the current dir. Sorry, if I'm missing something obvious, I know very little about C

Tested successfully on Windows 7 64 bits, compiled with libpng 1.6.3 and zlib 1.2.8.
Link: http://css-ig.net/tools/zip/pngquantv2-2.0.1.zip

Happens on Linux (compiled by someone else):

pngquant, version 1.7.2 (February 2012), by Greg Roelofs, Kornel Lesinski.
Compiled with SSE2 instructions.
Compiled with libpng 1.2.46; using libpng 1.2.46.
Compiled with zlib 1.2.3.4; using zlib 1.2.3.4.

and Windows (binary from the website):

pngquant, version 1.7.2 (February 2012), by Greg Roelofs, Kornel Lesinski.
Compiled with libpng 1.2.37; using libpng 1.2.37.
Compiled with zlib 1.2.3; using zlib 1.2.3.

I know, I know, why would you want to optimize such small PNGs? Well, they are part of a batch job of 1000s of differently sized PNGs.

I am having some ideas when I am compressing a picture mixed with text and photo, but I am not sure how feasible or reasonable it is for pngquant to have a feature for this. Text and photo require different level of quality in compression and most compression tools assume that we are working with only text or photos, so that requires quite a bit of work for me to crop pictures and recompose them after compression, which introduces new performance overheads. I wonder if it is possible for pngquant to determine approximately photos and text in a picture (through a pixel distribution histogram maybe) and compress them with different qualities.

Or if I can come up with bounds for photos and text in the picture, would it be easier for pngquant to selectively apply different compression setting for different bound box in the picture file?

Hi,

Kindly package future release of code inside top level directory:

PACKAGE-VERSION/

So that they can be unpackged nicely:

pngquant-version/
pngquant-version/
...

Unpacking now in wrong directory litters all other files in there.

Thanks

Hello, I use the pngquant on my css auto sprite tool: joycss. I find that, the old version 1.7.2 has a bug on IE6. The background color will be dark than the origin when a picture is processed by pngquant. When I update to the newest version, the problem is gone.

But, I do not know how to build it at win, build at mac or linux is easy. So, I wish you could help to do that, build the newest version. I think, the unix like system user must know how to update to the newest version, what most needed is win exe file.

thank you, pngqaunt is a great work.

CFLAGS line: "-std=c99" switch is missing

OBJS line: "pam.o" missing