Preface
Downloading the models
Unpacking the models
Apply fixes and patches to the models and store them
Create model index files
Test the models
Use the models with gEDA
Maintaining models
Getting, testing and maintaining spice models can be a very time consuming task. This is a proposal on how to automate or speedup some of the steps.
The second problem is that distributing models may not be allowed. (copyright notices on the vendor pages).
Thus I propose to distribute the building rules and not the model files itself. This building rules may be stored in and SCM tool somewhere in the web. Note: This is the way openSUSE distributes proprietary font packages, ...
Most of the steps below may be part of a SCons config file, others can be done with scripting.
Note: This project now uses SCons (http://www.scons.org) instead of Make.
The models can be downloaded with a download utility like wget, curl or something similar using a script language like python, ...
Each vendor needs an extra set of download instructions. The downloaded files are stored in a specified directory structure:
download/[vendor]/[files]
Downloading is splitted from all tasks below. Thus you can apply the steps below again and again.
Most vendors provide there models as .zip archives with different directory structure. In this step the models gets unpacked.
In this step all files gets a md5 checksum. The checksum files are stored in:
model_checksums/[vendor_category].md5sum
This checksum may help to track changes in the model files when maintaining the model library. You'll easily be able to find the changed files of a .zip file.
The model files from the vendors needs to be fixed and the fixed files needs to be stored in an usable directory structure. Possible fixes are:
The fixed model files are stored in a new directory structure:
model_library/[vendor]/[section]/[modelfiles]
I'm storing checksums of that created lib, too:
model_checksums/[vendor_category_lib].md5sum
The index files contains the information how to use the models and how to combine them with gEDA symbols. The information is stored in an .ini-style file format that can be easily written with scripts and with an editor. The index files are stored in:
indexfiles/[vendor_section].index
The index file has a GLOBAL section with various attributes:
[GLOBAL] MODELDIR=model_library/nxp/bipolar/ TESTDIR=model_tests/nxp/bipolar/
And symbol definitions:
[2PA1576Q_NXP00001] symbol=pnp.sym value=Q2PA1576Q modelname=Q2PA1576Q file=2PA1576Q.prm refdes=Q? pinseq_c=1 pinseq_b=2 pinseq_e=3 pinnr_c=C pinnr_b=B pinnr_e=E footprint=none test_refdes=Q1 model_status_good=ngspice_23 model_status_broken=gnucap_0.35
The section has a "name" part and a "uniq number" part. Thus all generated symbols will be uniq. (important for gschem, ...)
"symbol" contains a pointer to the symbol template file to use.
"value" to "footprint" are variables that will replace placeholders in the symbol template files.
"test_refdes" is currently required for testing.
The "model_status_good" is an optional, manual rating of a model. It's value is a list of all simulators that have been tested. With "model_status_bad" bad models can be marked.
It's important to get tested models. The tests are applied by a library testing script. The script:
With these steps the whole tool chain is tested for every model.
A human tester can watch the simulation result with a browser and doublecheck the simulation results. This results can also provide a quick view for the users of that model. The test result can be written to the model_status in the index file.
If the simulation fails, then either the description in the index file is wrong or there's something wrong with the model.
If the model is wrong you can either write a patch, fix it with a script in the config file or just mark the model as broken.
The new components can be used with gEDA after connecting it with a small script and the component-library-command backend.
Maintaining means that you walk throught all the steps in a fast way.
If the downloaded model archive is identical (checksum) to that you've already used, your done.
If the archive is different but contains 95% of identical files, then you only need to check the modified and the new models.