kicad-developers team mailing list archive
Mailing list archive
I noticed it was asked in the user list so I resuscitated it.
I did it to use 356 instead of gencad for board testing but in the end
the fabricator preferred gencad anyway...
However, here it is:
- Tested only with 3 boards (only dual side), so stuff may be wrong.
Files are read correctly by both CAM350 and (an old version of)
gerbtool. Real network names in the CAM are actually useful.
- Fixed function, no dialog. Simply asks for the output file.
I arbitrarily chose d356 as extension, another one found in the wild
- Always uses the auxiliary axis and always marks via as tented (I think
these could be the only options)
- Only does the old 356 format, *not* the 356A (however it seems the old
format is somewhat more supported); shouldn't be too difficult to
extend for the 356A format.
- Outputs in decimils (i.e. CUST 0 format); gerbtool ignores the unit
parameter so CUST 0 is the only one it works for it (it seems there is
a race in suckitude between CAM vendors...)
- It tries to be smart truncating network names, however there is no
check for other kind of string overflows (the refdes, in particular).
Coordinates however are clamped to the maximum representable (what
about bigger boards/features? there seems to be a scaling facility but
I guess nobody handle it correctly).
Extended characters in refdes/pin name could be a problem too (since
UTF would expand them and break the column alignment).
Also, the specs are not anymore available, I did it using some PDF
I found around; some things are not clear (so they may be wrong); in
- The mask specification seems to imply that the value specified
indicate the sides unavailable for probing; i.e. a tented via is 3 and
a THT pad is 0.
- Side number for access code: 0 is both sides, 1 is primary size (i.e.
top) but the other layers? on a 'finished board' there is only the
bottom side but the specs talk about inner layer too. So I export
top as 1, bottom as num_copper_layers and the internal are between.
I'm doubtful about the testability of inner layers before laminating,
- I don't remember if I did the locale switching to C thing during the save;
however it only writes integers so it shouldn't be an issue.
How it works: really simple... it looks over all the vias and all the
pins in the board and accumulates a vector with the info; then writes
the file using the vector (a bunch of printf). It could have written it
during the scan but later formats require a sorted vector, so I looked
forward to them. The only fancy thing is keeping a map of shortened
network names (one of the limitations of the format is only 14
characters for the netname)