Hi Sandor
I haven't used OpenSCAD but I thought it worth mentioning some personal
points about using Git for Engineering products, where variants
proliferate and build-products must be versioned.
I had a go with FreeCAD a few years ago to create an insulating spacer
that the mech eng had forgotten...
On 05/01/2021 09:57, Mr. Sandor Kunyik wrote:
Greetings,
If you were to design a new workflow, what key observations would you
make in regards to OpenSCAD scripts?
I'm sure that git is able to store and version all the scripts and
products, but it does need a vision about how to present the
organisation of the scripts and products, especially for engineered
items. Remember that software version control is just drawing office
procedure with the hard stuff ripped out ;-) [1]
As a quick illustration: I have a model with cavities to hold hex
bolts and nuts. I fine-tuned the model to print on PrinterA, using
FilamentA, SettingsA.
Is the fine tuning done via the scripts, or is it manual. If it is the
latter then it will need the user to record in the commit message for
the revision the various what/why/how aspects that aren't obvious from
the 'diff'. Does OpenSCAD have an informative 'diff' capability? If it
is just a 'compile for printerA using known-tool' then it could be an
automated commit message as you are storing a build product (If it was
regular software it's likely it's a binary..)
Once in "production" I need to "freeze" all relevant scripts,
especially when using multi-file structure. If the modules receive
parameters I need to "remember" those parameters (such as the radius
for the hex, and the dept of the cavity), and if they use hard-coded
values I must remember not to change them. Otherwise I cannot
repeatedly print the same model.
So these parameters are essentially a 'configuration file' (inc. the
hard coded values). If you auto generate such a file then it can easily
be stored within revision (commit) - It is important that the scripts
should be able to work directly from that config file (idempotent [2])
Now imagine this for the entire standard set of hex bolts - each of
these were fine-tuned, test-printed and verified.
A library? Hence a sub-module perhaps.
The rationale behind this to guarantee that the models trying to
conform to a standard (such as ASME B18.3) stay put, while models
receiving non-standardized sizes such as Nylon 6/6/ (which have bigger
hex heads) stay separated, and tweaked to work with each supplier.
A configurable library item? Tricky. I remember a project that need to
change 20,000 drawing because they changed the paint colour (green to
grey), and then the next major order was in the old colour! In software
terms its like 're-skinning' a GUI, but worse.
More likely you'll have a separate sub-module-library for these
'specials' where the source part is cherry-picked (give give a backward
textual reference) and then modified locally giving git level
traceability.
My question is, should I just "hardcode" everything, set up forks or
branches for all past scenarios? So far I only have a few dozen models
and I'm already having a hard time finding models I printed and used
in the past, to print again. How do I structure all this?
The hard part is to be both the 'design authority' that signs off the
'release' and also the day to day designer trying things where it's a
'fingers crossed' hope that this 3d print run will actually produce
what
you want and expect, but mainly it's a case of realising the mistake
and
cycle it around.
You should at least use tags for the good release commits, to make them
easy to find.
I am a mechanical engineer not a coder, new to all these.
I feel you pain. Some of the above may be more suited for those who
actually design and contribute OpenSCAD, rather that a user getting
started.
Maybe git or revision control is not the correct tool for this job?
As an ME you will know that without [good] revision control the world
falls apart for anything other than bespoke item home jobs!
Git is great because it give you full control, and local storage,
without taking up too much room.
Git is a problem because it gives you too much control and too much
choice. Start simple, be open to the 'mind-shift'.
Add any (i.e. more) files that you use to Git. Remove them (one at a
time) when you stop using them. They will still be in the history!
Commit often: including halfway through a change - you know you will
add
more and then more, then more tweaks before you print - have a commit
for each. I.e. distinguish between 'Printing' runs, and the act of
'visualising the part' in the OpenSCAD viewer - both steps get a
commit,
with concise description. (probably every 5 minutes or less!)
Use the 'git gui' and 'gitk' to visualise how you have developed and
progressed. (or a GUI of your choice -> CAD people are NOT terminal
people!!)
Git unfortunately lacks a short command to allow you to park a
completed
items onto a 'Completed' branch that's just an easy quick way of seeing
those finished items (Git has the view that they [old software
versions]
are 'gone/left behind', with just a few v# tags), so I don't have any
great solution for that.
If you have concise commit subject titles and messages, then it's
easier
to search history for "print: insulating spacer widget v2.1" (and
before
that "visualise: insulating spacer widget v2.1a", 1b, 1c, 1d, 1e...).
You may want to allow-empty commits for ease of recording!
Hope that helps.
[1] Drawing Office procedures allow for non perfect machining, serial
numbering, etc. Software generally asserts perfect replication,
repeatability and machining.
[2] https://en.wikipedia.org/wiki/Idempotence
[3] concise: adjective - giving a lot of information clearly and in a
few words; brief but comprehensive. "a concise account of the country's
history" (Git: without repeating the obvious changes shown in the
diff!)
