Comment by teamoet | Hacker News Reader

Comment by teamoet | Hacker News Reader

Parent Full thread

I've seen this fork mentioned around a couple of times so I've decided to read the articles on https://ansel.photos/en/news/. I'm not trying to deny that his motives are right, but the way and how often he bashes on darktable developers is really off-putting. I'll only cite a couple but they're easy to find:

> a handful of guys with more freetime and benevolence than actual skills

> So I fixed the whole logic [...] You might think that was a problem solved and a job well done, but that’s leaving Darktable’s geniuses out of the equation.

If you want to work alone I guess you can have that sort of negative attitude... but to me it clearly says "don't use or contribute to this software".

It's easier to spot mistakes after others have already made them, and then come up with better approaches. And it's easy to find yourself complaining about what is basically a prototype somebody else made and spared you the effort. It can definitely pump your ego up.

I'm saying this because one-man forks almost never lead to popular adoption, and almost always lead to abandoned forks, even if the new developer is technically gifted. I'm somewhat reminded of KWinFT (KDE fork) that has been somewhat recently renamed to Theseus' Ship.

I understand that you acquired a repulsion to design by committee, but when dealing with large projects you can't do it all by yourself, so you need to start learning how to deal with people. But who knows, maybe it's possible to find other like-minded contributors who are not so easily thrown off by the immaturities of a project's leader.

It's funny how when one guy does it for a small fork on a small project, people get riled up about it. However, when there's a bigger project like HomeAssistant, people let that behavior slide.

As far as Darktable/Ansel goes, he's right. Darktable's UI and philosophy is pretty horrid. I shouldn't have to know seven different algorithms to apply a denoise filter. The vast majority of professional photographers are artists, not computer scientists. I want the application to pick the best one for me. All commercial applications these days take it one step further with some pretty good AI tools, too.

Which creates shallow artists who just want it to "look pretty now." Not saying there isn't a place in the world for that, but there are always trade offs with tools that manipulate your human vision of art, and to say "there are too many denoise filters, just pick one for me" will be severely limiting when you realize that what you really wanted was grain removal, but that's not how your AI denoise filter works.

Again, there's room in the world for all manner of software uses. But to argue that Darktable is bad because it gives too many options, misses the goals of a great many artists, which is to understand what's happening to the pixels they captured in the field.

Auto modes have their places, as does retaining specificity of features. I don't think they're exclusive to each other.

And I don't think simple choices create shallow artists (or that the goal of Darktable is to create artists). Someone who doesn't have any arts education already doesn't have the technical understanding or vocabulary to really know what they're doing, so maintain the extra barrier? How many professional grade tools can you think of that have simple or guided modes?

The great part about software is that done well, it's often designed to be functional without a depth of specialty or expert knowledge, at least no more than a homeowner telling the builder "make my driveway to here" needs to know how to source and formulate concrete so that the end product looks good and doesn't crack or weather.

You're arguing that software makes or breaks artistry, and it doesn't. Composition, subject, and lighting still get you 90% there. Software just comes in at post and it plays less of a role than the HN crowd thinks it does.

Maybe it's my circles (photojournalists) but none of them care about different algorithms in post. Their artistry comes from a complete mastery of layers and masks along with old fashioned tone and color sliders. Those tools are far more powerful than knowing the difference between method 1 and method 2 of a Gaussian Blur filter that have no discernible visual difference anyway.

Hard agree - you probably aren't making great art from learning a different denoise algorithm, but you certainly are by learning composition skills. There are great photographers out there who barely even touch the adjustments in software. And when they are touching the adjustments, it's probably not so technical focused as choosing a different algorithm. It'll be "creative" adjustments like tone mapping and colour LUTs.

The problem with DT isn't just that it presents too many options, it's that the interface itself is bad and the options are presented poorly. Providing a million knobs to fine tune things is not a substitute for smarter, more coarse adjustments. The vast majority of end users don't want to have to manually set the TCP flags for each network request in a web browser. Darktable is better if you're wanking about with an academic paper, but it's just a plain bad experience for photo editing.

With your example, the multitude of denoise filters is appropriate because one of them is actually the grain removal. When in fact grain and noise are two separate issues.

How about demosiacing? There are nine different algorithms to choose from. Great if you're writing a research paper, useless if you're actually working on a photo.

How about white balance? If you thought that twiddling the knobs in the "white balance" was the way to go, boy are you in for a surprise. That'll just trigger warnings and errors. What you really wanted was "color calibration". You didn't want to change the color temperature, you wanted to worry about gamut compression and illuminant.

Now let's say you're working on a photo and have already narrowed down some settings you'd like to commonly apply. Well. First you have to drop out of the "darkroom" module and go to the "lighttable" module. Then you have to accept that your edit history is going to get fucked because darktable doesn't store a history of individual changes but rather aggregates them often by module. So if you hit undo, you potentially undo more than just the "style" you just applied. But what style did you just apply? Darktable (intentionally) doesn't keep that information around at all. You can have the style add an instance name, but Darktable eventually gets confused if you go back and forth between different presets.

To add insult to injury, there's no A/B view in Darktable. Nine different demosiacing algorithms but no easy way to compare them.

Darktable misses the forest for the trees.

Not to detract from the entirely justified criticism, but in case it helps somebody: The "Snapshot" feature in Darktable can be used to compare two renderings. It is a bit clumsy but this can be used to compare output of two algos.

That pretty well underscores my point, however. Darktable can do much of what Lightroom does, just in a more tedious manner. When confronted with this the DT devs respond with a "DT is just too powerful for most users". The biggest problem with DT isn't that it's too powerful it's that the interface is just dreadful.

If you like, or want to use Darktable/Ansel - check out Harry Durgin on YouTube.

He does a fantastic job of explaining how to use Darktable (and photo manipulation software in general) to get the best out of it.

https://www.youtube.com/channel/UCngiFhMSngeFwUGtsXA-17w

  but to me it clearly says "don't use or contribute to this software"

As an end user the sheer arrogance and condescending attitude that the darktable devs bring to the table is far more offputting than someone (an ex-darktable dev no less) losing patience with that behavior.

Still, there’s a lot to be said for taking the high road. If the attitude of the upstream devs is so toxic to drive you to create a fork, then why not differentiate yourself by creating a toxicity-free community to the best of your abilities?

You might even adopt forum rules similar to HN’s but with a focus on improving access to open source photographers’ tools (which is what this software is supposed to be in the first place).

Darktable has so thoroughly sucked the fun out of digital photography for me that I can hardly blame Aurélien for being salty. I just don't care that much that his frustration boiled over because he's at also doing something constructive.

Seconding chongli, you can be very frustrated with the devs and not be toxic; losing patience, even if justified, does not justify being an asshole, and even genius devs like Linus have grown to understand this. I agree with some points Aurélien makes and do find Darktable to be a bit frustrating to use, and I'm sure he's right about a lot of the reasons it's slow and not great to develop on, but he crosses a line, going from criticizing to belittling others and propping himself up at the same time, and just sounds like an insolent and narcissistic person, which makes me basically never want to try his fork.

Worth noting Aurélien was a Darktable dev, so he's not some outsider that's coming in like a wrecking ball. I think it's much to Linus' benefit that he's mellowed out over time. However I'm not going to be quick to condemn someone who's already contributed significantly to DT.

You're quite a bit less motivated than I am to try Ansel I suppose. I found DT (and the devs' defense of their decisions) intolerable. To the extent that I got Ansel built on my mac. Aurélien indeed fixed some of my biggest complaints with DT, but unfortunately he ripped out some bits I actually wanted to use.

At this point I suppose Ansel can go one of two ways: DragonFlyBSD or LibreSSL.

I'm a 100% with you on this, but I always try to be as neutral about this as I can. A character/temper of a developer should not be the main topic of discussion, the software should be though.

To your comment about committees. That (or the lack thereof) is a big critique point the ansel developer makes. A leadership of one is better than the leadership of none. And since he forks darktable and mostly removes and replaces some functions I don't see it as a problem in this case, especially since he has been maintaining the project for many years now.

Anyway like I said, I would recommend trying the software and if you think it's worse than darktable just don't use it :) I for instance have multiple software packages installed and am quite annoyed to have to use multiple packages for one "thing" and I always try to use the best tool for the job. On photo management/editing I'm quite torn

Except when you look at projects like FreeCAD, the lone developer forking the project (real thunder) has seemingly done more for the project than the original developers. A lot of this has to do with the fact that everything needs to be endlessly discussed on the FreeCAD forums before anything gets done. Considering the limited resources on the original developers hands, this is a recipe for stagnation and standstill. It is faster and simpler to just do the work and then merge the patches from the fork than to go through the official, slow and unproductive way.

Yeah sure very few people use realthunder's fork including me, but I have so far significantly benefitted from his work in the official FreeCAD release.

There are dozens of things that just need to be implemented and not pointlessly discussed in FreeCAD. The things I'm talking about are absurdly blatant and obvious to anyone. You know, things like a transparent preview of the operation you're doing. A sane attachment editor that doesn't choose a sketch orientation at random.

The negativity is necessary since the original developers are shutting valid criticism up with fake positivity.

Except they didn’t merge patches from RealThunder’s fork for TNP, for example. They used it as a guide and reference implementation and wrote a new implementation.

In general they do not just merge his work; they have to rewrite it. RealThunder is prolific but he evidently doesn’t use the same coding standards as the rest of the project, and makes changes across workbenches where he chooses for his own ideas, when in the core project they have other maintainers.

He has the total freedom to do this, and I agree his fork is illustrative of good solutions in some cases, but this is not a good way to just fix master. So they don’t.

Transparent previews in Part Design — and a general mechanism for them elsewhere - is coming in 1.1.

I notice you talking about the attachment editor choosing a random orientation a lot: in my experience it does not choose randomly, if you use an appropriate attachment scheme. I think I have rotated a sketch attachment once in my last two dozen or so uses, and that rotation was indicated by the design. The heuristic is complicated though, and the interface has several frustrations. There could be better UI for working through the attachment schemes.

1.1 has a change to core datums (Part Design-style LCS, datum planes, lines and points available throughout FreeCAD, not just in Part Design) that should make some of the more esoteric attachment schemes less often required, because you will be able to place an LCS once

Reply on news.ycombinator.com