Just want to get this on my list so I can refer back to it.technosaurus wrote:I reduced the sizes of some of the danpei xpm icons a bit (enough to save ~20kb in the compiled binary size)
I may as well describe what I did to shrink them
open with mtpaint
Image -> convert to rgb
Image -> convert to indexed
check Min Max Quantize and Floyd Steinberg
change the number at the top to 92 (maybe 93?) or less
save as whatever (works for at least xpm and png, possibly others)
(use your best judgement based on the image - lower number of colors reduces size some, but may end up looking like trash if it gets too low - most small icon pixmaps look ok at 16 colors)
this works because reducing the number of colors below 93 causes the pixels to be represented by a single character ... effectively cutting the size in half (not including the reduction for each color ... ~16B ea. I think)
Note: I have had issues with some versions of mtpaint with this method especially if you do further manipulations of the pallette - it will sometime lose a clear background and need to be edited in a text editor.
pUPnGO - 6Mb ISO - Basic Building Block Puplet
For console applications daily reviews and tests...continues the path of http://kmandla.wordpress.com/
Dragging up the original pupngo thread to continue blabla on more basic things...
Back in tinyXlib testing - got stuck with xdesktopwaves. It compiles ok but refuse to run as it cant translate color #406080 and other colors in same syntax. Adding "64 96 128 #406080" to rgb.txt helps a lot but seems to be an unusual hack.
So updated GetColor.c, Xcmsint.h, LookupCol.c, ParseCol.c. Added cmsCmap.c, Cmap.h, Cv.h, CCC.c, cmsColNm.c, XRGB.c, cmsInt.c, cmsGlobls.c, CvColW.c, LRGB.c, CvCols.c, cmsProp.c, HVCGcC.c, XYZ.c, uvY.c, xyY.c, Lab.c, Luv.c, HVC.c, cmsTrig.c, cmsMath.c, AddDIC.c, HVCMxC.c, HVCMxVC.c to X11lib which solves the problem.
BUT adds 20-30K to every binary linked against the new libX11. And most applications does not use the added functions at all. I do compile with "-Wl,--gc-sections,--sort-common,-s" but seems that the new functions are "glued" in...
SO: Are there any way to map the internal dependency of the functions in a library from source-code? If so this could be used to evaluate the consequences of adding functionality or to put in conditions when to use and when not to use...
The obvious solution to xdesktopwaves is to patch the source to use names from rgb.txt and bypass all above additions - but there might be applications that really benefit from the added functions...
Back in tinyXlib testing - got stuck with xdesktopwaves. It compiles ok but refuse to run as it cant translate color #406080 and other colors in same syntax. Adding "64 96 128 #406080" to rgb.txt helps a lot but seems to be an unusual hack.
So updated GetColor.c, Xcmsint.h, LookupCol.c, ParseCol.c. Added cmsCmap.c, Cmap.h, Cv.h, CCC.c, cmsColNm.c, XRGB.c, cmsInt.c, cmsGlobls.c, CvColW.c, LRGB.c, CvCols.c, cmsProp.c, HVCGcC.c, XYZ.c, uvY.c, xyY.c, Lab.c, Luv.c, HVC.c, cmsTrig.c, cmsMath.c, AddDIC.c, HVCMxC.c, HVCMxVC.c to X11lib which solves the problem.
BUT adds 20-30K to every binary linked against the new libX11. And most applications does not use the added functions at all. I do compile with "-Wl,--gc-sections,--sort-common,-s" but seems that the new functions are "glued" in...
SO: Are there any way to map the internal dependency of the functions in a library from source-code? If so this could be used to evaluate the consequences of adding functionality or to put in conditions when to use and when not to use...
The obvious solution to xdesktopwaves is to patch the source to use names from rgb.txt and bypass all above additions - but there might be applications that really benefit from the added functions...
where you been man? I can't imagine I'm the only user (you're a dev, I'm a user 
) who cares about pUPnGO...
-
technosaurus
- Posts: 4853
- Joined: Mon 19 May 2008, 01:24
- Location: Blue Springs, MO
- Contact:
Re: added functions.
If you are talking shared lib size, you can include uncommon functions in the .h instead of the .c, but the gotcha is you need to ifndef it in the .h so you can define something like -DBUILDLIB to exclude it, that way future builds will build a static object and hopefully ffunction/gc-sections will pull it from bulds that dont need it.
If youve ever looked in a config.h (and how it is generated) it may be clear using the HAVE_/NEED_ defs as examples ... I think jwm has a shell version.
This is actually an interesting approach to overall shared lib size reduction in general and could be extended to easily inlined small functions that consist of a single conditional + a return ... Since they arent included in the .so, it shrinks and since the inlined code is smaller than a funtion call + return for these really basic fxns.
Youd be surprised how often a giant series of ifelses, can be reduced to a ternary op
return ( A || B && C)? -1 : 0:
Where the A,B,C expression is the eqivalent of the ifelse.
If you are talking shared lib size, you can include uncommon functions in the .h instead of the .c, but the gotcha is you need to ifndef it in the .h so you can define something like -DBUILDLIB to exclude it, that way future builds will build a static object and hopefully ffunction/gc-sections will pull it from bulds that dont need it.
If youve ever looked in a config.h (and how it is generated) it may be clear using the HAVE_/NEED_ defs as examples ... I think jwm has a shell version.
This is actually an interesting approach to overall shared lib size reduction in general and could be extended to easily inlined small functions that consist of a single conditional + a return ... Since they arent included in the .so, it shrinks and since the inlined code is smaller than a funtion call + return for these really basic fxns.
Youd be surprised how often a giant series of ifelses, can be reduced to a ternary op
return ( A || B && C)? -1 : 0:
Where the A,B,C expression is the eqivalent of the ifelse.
Check out my [url=https://github.com/technosaurus]github repositories[/url]. I may eventually get around to updating my [url=http://bashismal.blogspot.com]blogspot[/url].
-
technosaurus
- Posts: 4853
- Joined: Mon 19 May 2008, 01:24
- Location: Blue Springs, MO
- Contact:
Re: added functions.
If you are talking shared lib size, you can include uncommon functions in the .h instead of the .c, but the gotcha is you need to ifndef it in the .h so you can define something like -DBUILDLIB to exclude it, that way future builds will build a static object and hopefully ffunction/gc-sections will pull it from bulds that dont need it.
If youve ever looked in a config.h (and how it is generated) it may be clear using the HAVE_/NEED_ defs as examples ... I think jwm has a shell version.
This is actually an interesting approach to overall shared lib size reduction in general and could be extended to easily inlined small functions that consist of a single conditional + a return ... Since they arent included in the .so, it shrinks and since the inlined code is smaller than a funtion call + return for these really basic fxns.
Youd be surprised how often a giant series of ifelses, can be reduced to a ternary op
return ( A || B && C)? -1 : 0:
Where the A,B,C expression is the eqivalent of the ifelse.
If you are talking shared lib size, you can include uncommon functions in the .h instead of the .c, but the gotcha is you need to ifndef it in the .h so you can define something like -DBUILDLIB to exclude it, that way future builds will build a static object and hopefully ffunction/gc-sections will pull it from bulds that dont need it.
If youve ever looked in a config.h (and how it is generated) it may be clear using the HAVE_/NEED_ defs as examples ... I think jwm has a shell version.
This is actually an interesting approach to overall shared lib size reduction in general and could be extended to easily inlined small functions that consist of a single conditional + a return ... Since they arent included in the .so, it shrinks and since the inlined code is smaller than a funtion call + return for these really basic fxns.
Youd be surprised how often a giant series of ifelses, can be reduced to a ternary op
return ( A || B && C)? -1 : 0:
Where the A,B,C expression is the eqivalent of the ifelse.
Check out my [url=https://github.com/technosaurus]github repositories[/url]. I may eventually get around to updating my [url=http://bashismal.blogspot.com]blogspot[/url].
Thanks! The permanent size increase is in static linked apps - I havent tested if the size of the dynamic linked also is permanently larger.
In the actual xlib case there seems to be 3-4 entries that triggers the added size: In GetColor.c, LookupCol.c and ParseCol.c we have a structure like below:
where the #ifdefs are mine. If included in the lib the size-increase is there even though the functionality is not used by the application - but the function "XLookupColor" is used so it drags the rest in...
It looks like an extension to XlookupColor as the content include this statement
In the actual xlib case there seems to be 3-4 entries that triggers the added size: In GetColor.c, LookupCol.c and ParseCol.c we have a structure like below:
Code: Select all
Status
XLookupColor (
register Display *dpy,
Colormap cmap,
_Xconst char *spec,
XColor *def,
XColor *scr)
{
register int n;
xLookupColorReply reply;
register xLookupColorReq *req;
#ifdef USECMS
XcmsCCC ccc;
XcmsColor cmsColor_exact;
/* more code removed */
#endifIt looks like an extension to XlookupColor as the content include this statement
Code: Select all
/*
* Let's Attempt to use Xcms and i18n approach to Parse Color
*/-
technosaurus
- Posts: 4853
- Joined: Mon 19 May 2008, 01:24
- Location: Blue Springs, MO
- Contact:
I wonder what is i18n about color lookup? if it is localized color names (never seen them)? can an lgettext -> C locale be a sufficient replacement? I am tempted to convert an rgb.txt as a switch case in C similar to my file mimetype code (it will add some size internally, but could speed things up and not require an rgb.txt)
Check out my [url=https://github.com/technosaurus]github repositories[/url]. I may eventually get around to updating my [url=http://bashismal.blogspot.com]blogspot[/url].
I am not sure but rgb.txt might be case sensitive - at least the names can be done in many ways:
At least icewm (or is it imlib) benefits from the added xcms functions - maybe mostly because the format #xxyyzz is widely used in icewm-themes - but I still cant see a way to conditionally use or exclude the xcms.
After playing with tinyX11 + MESA + freeglut doing fvs & tdfsb a wish for making small videos of screen-events popped up...
After testing recordmydesktop (which build static but refuses to run) I went through various versions of xvidcap. The newer versions seems quite gnome dependent but the older 1.0.XX version builds ok and run.
Last version is xvidcap-1.0.19 and a static build is attached.
For creating a video external programs are still needed. Mencoder or ffmpeg can be used (transcode as well). ppm2mpeg is a script which works well with xvidcap and I have included it in the attached pet.
For editing, tools from ImageMagick are called but they are HUGE when static linked.
A quote from the xvidcap README:
Code: Select all
248 248 255 ghost white
248 248 255 GhostWhite
245 245 245 white smoke
245 245 245 WhiteSmokeAfter playing with tinyX11 + MESA + freeglut doing fvs & tdfsb a wish for making small videos of screen-events popped up...
After testing recordmydesktop (which build static but refuses to run) I went through various versions of xvidcap. The newer versions seems quite gnome dependent but the older 1.0.XX version builds ok and run.
Last version is xvidcap-1.0.19 and a static build is attached.
For creating a video external programs are still needed. Mencoder or ffmpeg can be used (transcode as well). ppm2mpeg is a script which works well with xvidcap and I have included it in the attached pet.
For editing, tools from ImageMagick are called but they are HUGE when static linked.
A quote from the xvidcap README:
This program does not use a special hardware driver to access the
video card. It just asks the X server about rectangular areas. This
means you need a fast machine (>= 133 MHz) and a fast harddrive. Big
frames (e.g. 384x288 = 1/2 PAL) at a high FPS rate are only possible
with very very fast systems
amigo: Sources are here
Needed below small patch to be able to compile:
Needed below small patch to be able to compile:
Code: Select all
--- ./src/mngutil.c.01 2003-10-13 22:53:52.000000000 +0800
+++ ./src/mngutil.c 2013-09-02 17:01:15.000000000 +0800
@@ -30,13 +30,13 @@
#include <png.h>
#include "mngutil.h"
-extern void png_write_data (png_structp png_ptr,
+/*extern void png_write_data (png_structp png_ptr,
png_bytep data,png_size_t length);
extern void png_save_uint_32 (png_bytep buf, png_uint_32 i);
extern void png_save_uint_16 (png_bytep buf, png_uint_16 i);
extern void png_write_chunk (png_structp png_ptr,
png_bytep chunk_name, png_bytep data, png_size_t length);
-
+*/
png_byte mng_sig[8] = {138, 77, 78, 71, 13, 10, 26, 10};
/*-
technosaurus
- Posts: 4853
- Joined: Mon 19 May 2008, 01:24
- Location: Blue Springs, MO
- Contact:
Yep, but when you strip the white space from and tolower the requested color, they are == (except for gray!=grey , but switch-case has fall through)... need to figure out which function(s) look at rgb.txt to figure out a good return format thoughgoingnuts wrote:I am not sure but rgb.txt might be case sensitive - at least the names can be done in many ways:
Check out my [url=https://github.com/technosaurus]github repositories[/url]. I may eventually get around to updating my [url=http://bashismal.blogspot.com]blogspot[/url].
@technosaurus: Some scripts and application reads rgb.txt direct - if not present they might fail...
I am trying to get a working version of ImageMagick - I have static builds of 3.6, 4.2.9, 5.2.2 and 6.8.6-9 - increasing in size from 700Kb to 4400Kb for "display". But running "display" is unstable with all versions. Even normal build in P412 linked dynamic is not stable. Xvesa/Xorg various settings does not help.
Problem is that the command menu works maybe every second time. Described here as well. Anyone knows more about this and how to solve?
I am trying to get a working version of ImageMagick - I have static builds of 3.6, 4.2.9, 5.2.2 and 6.8.6-9 - increasing in size from 700Kb to 4400Kb for "display". But running "display" is unstable with all versions. Even normal build in P412 linked dynamic is not stable. Xvesa/Xorg various settings does not help.
Problem is that the command menu works maybe every second time. Described here as well. Anyone knows more about this and how to solve?
-
technosaurus
- Posts: 4853
- Joined: Mon 19 May 2008, 01:24
- Location: Blue Springs, MO
- Contact:
should be easy enough to patch those, but I may take a look in the xserver code to see what it uses to send the colors, I've been meaning to see how hard it would be to port xfbdev to xcb anyhowgoingnuts wrote:@technosaurus: Some scripts and application reads rgb.txt direct - if not present they might fail...
unstable how? doesn't "display" just do the same thing as meh/feh with support for a few extra formats?I am trying to get a working version of ImageMagick - I have static builds of 3.6, 4.2.9, 5.2.2 and 6.8.6-9 - increasing in size from 700Kb to 4400Kb for "display". But running "display" is unstable with all versions.
Check out my [url=https://github.com/technosaurus]github repositories[/url]. I may eventually get around to updating my [url=http://bashismal.blogspot.com]blogspot[/url].
ImageMagick display can do image editing (annotations, oilpaint, sharpen etc) ...leftclick in the imagewindow and a menu pops up with all the possibilities. Unstable in the sense that the menu pops up but only "now-and-then" you can use it.
Further tests shows that icewm, xfce and fvwm do not give the problem so its likely to be a jwm-issue...
Attached quick implementation (from jwm commit da207c3) fix it but I don't like patching jwm for a problem only seen in ImageMagick...
Found hints to jwm here
Slight update: The quick patch does not solve everything - haven't expected that anyway. The jwm-762 solves everything concerning ImageMagic but adds unwanted stuff too: Apps writing to root background cant be seen anymore and the sticky windows are back in pager. Besides that is seems to run ok. Attached a static linked version of jwm-762.
I would have liked to patch ImageMagic instead but cant find my way in its code. So I will let it rest as is for now. Version 4.2.9 seems to do more than just the basic things needed to work with xvidcap-1.0.19 so made a multi call binary of all the tools (animate combine convert display identify import mogrify montage xtp) which ends up 1391Kb linked static and 517Kb upxed. Not small enough to post here but I have created a pet-package for download ImageMagick-4.2.9.pet here [591K].
Also attached patch used to facilitate the mcb-build.
Note: Build without tiff-support...
Further tests shows that icewm, xfce and fvwm do not give the problem so its likely to be a jwm-issue...
Attached quick implementation (from jwm commit da207c3) fix it but I don't like patching jwm for a problem only seen in ImageMagick...
Found hints to jwm here
Slight update: The quick patch does not solve everything - haven't expected that anyway. The jwm-762 solves everything concerning ImageMagic but adds unwanted stuff too: Apps writing to root background cant be seen anymore and the sticky windows are back in pager. Besides that is seems to run ok. Attached a static linked version of jwm-762.
I would have liked to patch ImageMagic instead but cant find my way in its code. So I will let it rest as is for now. Version 4.2.9 seems to do more than just the basic things needed to work with xvidcap-1.0.19 so made a multi call binary of all the tools (animate combine convert display identify import mogrify montage xtp) which ends up 1391Kb linked static and 517Kb upxed. Not small enough to post here but I have created a pet-package for download ImageMagick-4.2.9.pet here [591K].
Also attached patch used to facilitate the mcb-build.
Note: Build without tiff-support...
- Attachments
-
- ImageMagic.png
- ImageMagic-4.2.9 running in exp-pupngo with jwm-762
- (121.47 KiB) Downloaded 821 times
-
- ImageMagick-4.2.9_mcb01.diff.gz
- fake .gz
- (5.65 KiB) Downloaded 295 times
-
- jwm.gz
- fake .gz - static linked jwm-762
- (192.31 KiB) Downloaded 294 times
-
- jwm_2.0.1_IM_try.diff.gz
- (5.79 KiB) Downloaded 286 times
Further blah on xvidcap...
Seems that Xt is used alone until version 1.0.19. After that gtk2 is introduced but Xt is kept until version 1.1.3 and removed from 1.1.4 and from there and on becomes more gnomish.
I did not succeed embedding ffmpeg in 1.0.19 but can do it with 1.1.3. Choose not to as it seems better to be able to choose which version of ffmpeg we want for other usage. A static linked version of xvidcap-1.1.3 with a slightly modified ppm2mpeg +a static linked ffmpeg-0.4.8 and its corresponding ffplay + ImageMagic-4.2.9 can be downloaded here - [xvidcap-1.1.3_plus_ffmpeg_0.4.8_plus_IM-4.2.9.pet/2Mb]
Seems that Xt is used alone until version 1.0.19. After that gtk2 is introduced but Xt is kept until version 1.1.3 and removed from 1.1.4 and from there and on becomes more gnomish.
I did not succeed embedding ffmpeg in 1.0.19 but can do it with 1.1.3. Choose not to as it seems better to be able to choose which version of ffmpeg we want for other usage. A static linked version of xvidcap-1.1.3 with a slightly modified ppm2mpeg +a static linked ffmpeg-0.4.8 and its corresponding ffplay + ImageMagic-4.2.9 can be downloaded here - [xvidcap-1.1.3_plus_ffmpeg_0.4.8_plus_IM-4.2.9.pet/2Mb]
I wasn't able to compile either 1.0.19 or 1.1.3 -even with your png patch. Are you probably using png12? I have png14 here after finding and patching several progs/libs which were written for png12.
I finaly got 1.0.19 compiling by deleting the mng_write_sig from xtomng.c. The program runs but i don't know yet if it works.
After a simple './configure' command, here's the errors when trying to build 1.1.3pl1:
I finaly got 1.0.19 compiling by deleting the mng_write_sig from xtomng.c. The program runs but i don't know yet if it works.
After a simple './configure' command, here's the errors when trying to build 1.1.3pl1:
Code: Select all
Making all in src
make[2]: Entering directory `/home/amigo/Downloads/xvidcap/xvidcap-1.1.3-p7/src'
if gcc -DHAVE_CONFIG_H -I. -I. -I.. -DVERSION=\"1.1.3\" -g -O2 -MT xvidcap-mngutil.o -MD -MP -MF ".deps/xvidcap-mngutil.Tpo" \
-c -o xvidcap-mngutil.o `test -f 'mngutil.c' || echo './'`mngutil.c; \
then mv -f ".deps/xvidcap-mngutil.Tpo" ".deps/xvidcap-mngutil.Po"; \
else rm -f ".deps/xvidcap-mngutil.Tpo"; exit 1; \
fi
mngutil.c:36:13: error: conflicting types for 'png_save_uint_16'
/usr/include/png.h:2823:1: note: previous declaration of 'png_save_uint_16' was here
mngutil.c: In function 'mng_write_sig':
mngutil.c:49:2: warning: 'sig_bytes' is deprecated (declared at /usr/include/png.h:1255)
mngutil.c:50:3: warning: 'sig_bytes' is deprecated (declared at /usr/include/png.h:1255)
make[2]: *** [xvidcap-mngutil.o] Error 1
make[2]: Leaving directory `/home/amigo/Downloads/xvidcap/xvidcap-1.1.3-p7/src'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/amigo/Downloads/xvidcap/xvidcap-1.1.3-p7'
make: *** [all] Error 2
-
technosaurus
- Posts: 4853
- Joined: Mon 19 May 2008, 01:24
- Location: Blue Springs, MO
- Contact:
I assume one was unsigned short and the other an equivalent typedef'd uint typeamigo wrote:Code: Select all
mngutil.c:36:13: error: conflicting types for 'png_save_uint_16' /usr/include/png.h:2823:1: note: previous declaration of 'png_save_uint_16' was here
BTW, I'm taking a grad school class where I have to pick an open source project to do "non-programming" work for. I have it narrowed down to 1 of:
libxcb (mostly documentation )
netsurf/nsfb (testing, bug reporting and documentation)
musl-libc (documentation, testing)
toybox (documentation, testing)
(any other projects I'm interested in would involve code contributions)
any feedback? I'm leaning toward xcb because it lacks the documentation required to port major toolkits from x11 ... perhaps a replacement for tinyX11?
Check out my [url=https://github.com/technosaurus]github repositories[/url]. I may eventually get around to updating my [url=http://bashismal.blogspot.com]blogspot[/url].