kicad-developers team mailing list archive

[Vesa Solonen <vsolonen@...>]

> > decision where to go is a bit difficult. My bets are on Cairo/Clutter
> > which would mean wxGC and possibly some direct access to native context.
>
> Oh pleeeeeasee... Cairo is slow as hell... gerbv is nearly unusable even without antialiasing, for example.

Oh yes it is, but only because it is currently fully sw rendered. As long 
as every primitive is drawn also without any caching it's going to be 
slow.

> Maybe when we'll all have FireGL or Quadros in our boxes :P:P current GPU are optimized in filling texture triangles,
> not drawing lines... cad application are notorius for being geometry limited (games are usually fill limited because
> its all pre-computed, usually).
>
> For PCB at least the alpha blending (or bitplane processing) could be usefully accelerated but remember than under
> OpenGL every frame you need to redraw EVERYTHING (short of using dirty tricks like rendering all the 'static' stuff
> in a pbuffer saving the RGBAZ values before rendering the dynamic one... but it only works if the dynamic stuff is
> not blended AND you need a rather spiffy card to freeze the whole framebuffer in that way). OTOH my old FireGL4
> blasted antialiased LINES on the screen like nothing short of a consumer card of MANY years after... (it had a
> dedicated geometry processor aside the main GPU :D)

Do you have any clue where the limit is? If you test Kicadocaml you'll get 
some idea where your HW is. I also did some testing for lines along the 
example here: http://www.siafoo.net/snippet/97 by using random vertex 
points in mouse event and was really surprised how many AAlines r200 can 
do.

Isn't vertex buffer objects and frame buffer objects supposed to remove 
command replay? Clutter makes canvas based opengl drawing possible, so 
much nicer to work with than raw opengl. Apples developer library seems 
to be a good place to look for hints.

http://developer.apple.com/graphicsimaging/opengl/

> I fear the day when I'll have to rework a BGA :(:( Leadless packages are already a discrete PITA :P

I'm actively avoiding it :) I think for prototype and hobby use, mixed 
smd/through hole is the most convenient. Less holes the better, etc...

> As for coding I feel C++ is too bloated (cmake takes forever looking dependencies... why???)... maybe because I
> program the PIC18FJ straight in asm :P (with many other architectures... for every project a different MCU :(((

AVR assembly was my first programming language and I still prefer low 
level stuff over anythin else.

> Do you mean that sticky stuff involving colloidal palladium? It's somebody else's work :P I get to insult them when
> they do the hot air leveling wrong and the holes fills up with solder, anyway :D

Thereabouts. Though I've been making colloidal Pd on site with previous 
tin reducer dip. I bet it pays to outsource PCB making any day on 
commercial basis. It seems just too difficult to get eutectic pads these 
days and I had to plate some myself for one project which definitely 
didn't want any whiskers... 
:D you use leadless hasl? It flows so badly that it seems to be the 
invitation for plugged holes. Not to mention http://nepp.nasa.gov/WHISKER/

> Oh, and RF is SICK :D I prefer stronger stuff in low frequency (the latest PCB I designed had some traces of over 300
> mil and a 70A relay soldered on it... BTW the 'oblong hole' feature works as advertised, the pcb factory read the
> files without troubles)

I guess it's up to definition what is RF. Some say nothing below GHz, 
but I'd like to say anything that burns your finger with dielectric 
heating :D I guess you had a bit thicker than 35u copper on it too? SMPS 
stuff or car audio or classified?

> I put a demo pdf in the file section... my webmail isn't working and I don't know how to attach to the mail with
> yahoo groups...

Heck, do I have to get Y account now?

-Vesa