Hacker News Reader

Comment by emergentstate | Hacker News Reader

I have pretty mixed feelings about FreeCAD, having used it for 100+ hours across a few different projects.

It’s wonderful to have a open-source alternative to the expensive CAD software that muscled its way into a near monopoly in academia/industry, but it suffers from some really inexcusable design problems, on a fundamental level. As it is, I don’t think it’s really usable as a serious tool for creating complex assemblies.

I’d be curious if anyone here has managed to use it effectively for a large project (something real-world, not just showcasing FreeCAD).

View on HN

> some really inexcusable design problems, on a fundamental level

Can you be more specific?

I understand that the Topological Naming Problem is big (discussed in https://www.youtube.com/watch?v=QSsVFu929jo), but as I understand it this is being fixed in the RealThunder branch (https://bit.ly/3iBzQly), and fixes may be flowing from this into FreeCAD 0.2.

My complaint is that it is extremely modal and as a result not very friendly, not intuitive even to a relatively experienced CAD user.

I’ve used a variety of CAD and it just really didn’t click. I know it’s open source, so not going to be as UX oriented, but still.

I would love to be proved wrong. If there’s something I’m missing please share.

I've experienced that user unfriendliness too. I'm not at all an expert in this, I'm a FreeCAD newbie that has been trying to decide whether to invest in learning FreeCAD or not.

I keep wishing that FreeCAD had the UI polish of Blender, and ask myself if Blender can do enough of the CAD workflows that it makes sense to learn Blender instead of FreeCAD.

I keep reading that technically speaking, mesh-based modelers are woefully insufficient for CAD tasks. But then I see videos like this, which suggest that Blender can handle most modeling tasks, with the main exception of bevels and chamfers: https://www.youtube.com/watch?v=n5SehCbhxFs

Mesh modeling in blender has similar downfalls to OpenSCAD. When you’re designing a bit of precision equipment (or what have you) you don’t want your circles turned into high degree polygons.

Can a lathe turn a polygon? Can a drill drill a polygonal hole? Maybe you can machine a polygon on a fancy CNC lathe, but you certainly wouldn’t do so for a turbine shaft.

Also for less precise things, say construction, you care about dimensions. Select an arc to get it’s radius for instance.

This is why design for traditional construction techniques best done in boundary representations: https://en.m.wikipedia.org/wiki/Boundary_representation

I will concede that one can easily create nasty insoluble problems with boundary representations if not familiar with the limitations of splines, but this only really is a problem with curvy things like ship’s hulls and aircraft.

There is no downside in my opinion to learning a bit of OpenSCAD, FreeCAD, and Blender and reaching for the right tool when you need it. Personally, I have found the UI/UX of FreeCAD painful to use, but it is sufficient for most of my CAD needs. And after Fusion 360 crippled their hobbyist licensing, I don't really want to get in bed with any of the non-free CAD tools.

I found OpenSCAD to be very friendly for parametric designs, but after a while I switched to the SolidPython wrapper as it was more flexible.

However, I'm hunting for a better way to create 3d models via code (Blender Python?) because 'extrude_along_path'->SolidPython->OpenSCAD->stl export simply could not handle my model.

OpenSCAD is great in some ways, but the fact that everything gets boils straight to triangles drives me nuts. If there were an OpenSCAD equivalent to work with boundary representation models I’d have found my CAD nirvana.

Would this fit the bill?

https://solvespace.com/index.pl

ZenCad?

https://mirmik.github.io/zencad/en/index.html

CadQuery should be better as it is using the OpenCASCADE cad kernel (same as FreeCAD) instead of a simplistic ad-hoc thingy.

Thanks for pointing this out! Looks pretty useful, will give it a shot.

I honestly think CadQuery has more promise as a way to get open source 3D components and modules to the mainstream, because it's fundamentally code, and can be reasoned about as such (diffed, merged, blamed, parameterised, modularised, reused and tested). Putting SolidWorks or FreeCad file into source control is a very different thing.

Also, as it's natively Python, you have all sorts of library support, up to and including things like numpy.

On the other hand, sketching with constraints is very intuitive. Making interesting shapes that aren't trivial CSG demos by fiddling with code just doesn't sound that great.

The future dream mCAD for me is something where extrusions and fillets and other 3d stuff is code, but sketches are GUI edited.

I wouldn't consider OpenSCAD real CAD. It doesn't aid you in design really, it's just a language to represent the design you already have in your head.

Without constraints, you're still doing a lot of stuff directly rather than declaratively, and without a GUI you're mostly limited to geometry you can imagine, because doing anything by trial and error in OpenSCAD takes forever.

The whole thing feels very open loop, like it's meant to be "imagine it>reason about it> code it" rather than "vauge idea > iteration > verify"

I could see it being useful in some cases for high level professional designers though.

The one thing I lack in OpenSCAD is dimensions: there is no good way to add any.

Did you find the Part Design workflow? It took a while for Freecad to click for me but it ends up being fairly similar to other CAD programs in terms of workflow.

A fundamental problem is that OpenCASCADE, despite being the most advanced open source CAD kernel, does kinda suck. It's single-threaded (FreeCAD has it on the main thread so any long operation blocks the entire UI) and it's… rather buggy.

Recent fun example I've encountered: doing a fillet in two places at the same time can result in either an unhelpful "resulting shape is invalid" or a crash inside OpenCASCADE, depending on the selection order (!) — https://github.com/realthunder/FreeCAD/issues/202

I have noticed that fillets in particular are a weak point for FreeCAD.

Topological naming issue is the biggest problem that they should address immediately. Otherwise they can’t call it a parametric drawing tool. Period.

I discovered the issue in a middle of a design where I tried to change what I thought was a parametric piece. It unrecoverable blew up everything.

I kind of lost trust in it after that. I ended up making a backup for every step on top of the regular backups, like: piece_name-step-1… etc.

I discovered the real thunder branch too late.

Maybe they should just scrap everything, make RealThunder lead dev, and start over from where he is...

I'm sure there's some great stuff in mainline though that deserves merging.

Nothing large(I'm not a mechanical guy, this is all incidental work) but I use the RealThunder branch at my job, and I'm building some furniture that I don't think I could ever have done without the software.

I've never used any other CAD though, so my evaluation might not mean much, to me, just having CAD at all completely changes everything, regardless of whether it's good or not, when your reference point is LibreOffice Draw and pen+paper.

Personally I also ran into some inexcusable bugs on my first try of FreeCAD recently. Sadly I don't have the link anymore, but the problem was offsets being incorrectly calculated in certain cases, and I found it to be a long standing bug in their issue tracker. There wasn't an easy way to work around it so I had to look elsewhere. Too bad, because FreeCAD did seem quite fine otherwise (Note: I never used SolidWorks, and my demands are amateur level).

For programmers I highly recommend OpenSCAD or Cadquery, which let you write code to build your models instead of using GUI. With this approach, "parametric" comes naturally, and so you don't need to learn a special constraint system, it's just your code (and a bit of math).

Cadquery was a bit hard to get started with being essentially a bunch of python scripts, but it's also more versatile and more powerful than OpenSCAD, and has a good collection of examples in their docs. OpenSCAD is still great for e.g. building simple models for 3D printing.

Even for simple models I've found OpenSCAD lacking. In 3d printing it is quite common to add chamfers and fillets to smooth objects. This is an extraordinarily amount of work in OpenSCAD, let alone the render times that increase significantly.

I agree. I'm working on a project to finish my basement, and decided to use FreeCAD for it. It mostly worked, but did weird things often enough it was frustrating. Often the solution was to look up on the forums for a hack to do things in a certain order and it would work. Sometimes there was no solution, like trying to print a floor plan of my 3d model always did weird things with the scale and the labels. I could mostly get it to "work", but really lacked polish that made using it more of a chore than I'd like, and I don't feel like I got the full power out of using it.

Curious if you have compared to SolveSpace?

Wow, that looks like what I may be after!

Last time I used SolveSpace, what I hated the most was how the undo button was near useless.. and https://github.com/solvespace/solvespace/issues/91 is still open :/

Undo/redo works great and now has a 100 step history.

I will give the latest version another go.

I last used it 3 years ago, and I agree, I found it difficult to use, and worse, it would frequently corrupt the model, or confuse the object hierarchy to the point I would have to start again.

Reply on news.ycombinator.com