FPC releases and A.I

[440bx]

@Graeme,

I find your approach to be very rational. 

Personally, I believe that a program should be kept close to "release ready" at all times.  That means when a feature is formally added, it means it is also extensively tested and nothing else is done until it, at least, seems to be bug-free.  Only after the program, with the new feature, is determined to be solid and functional does the process move on to adding a new feature (or making any other change.)

[valdir.marcos]

... and nobody can force you to release FPC 3.2.4. Yes OK, I truly hope somebody will fork FPC project and starts to maintain it with a better attitude.

If the "usual" process takes too much effort for the FPC team, then why not reduce the scope. What's the most popular platforms? Window & Linux?

So then why not limit the scope to making "official releases" that target 90% of FPC users. Users of other platforms will have to follow the instructions on compiling their own compiler and RTL [it's really not hard].

Disclaimer: This suggestion doesn't mean the FPC team need to skimp on quality. I'm assuming there are automated tests that run nightly using VMs to cover all or most platforms FPC supports. Making sure the "fixes" and "main" branches are functional.

And with reduced scope there might even be time to turn releases into a "train" release schedule. Time based, not feature based.

eg: a 6 or 12 month release schedule with a release-branch freeze 1 (maybe 2) months before a release. If a feature is ready - it's in the next release. If it wasn't, it'll catch the next train, and be in the release after that.

With fpGUI I was also stuck in the rut of "feature" releases, and I had constant issues with feature creep. It had to be perfect, and what about just adding this or than extra feature... In the end it took years to make a release and I got less useful feedback.

I got over that, and now switched to time based releases. It gets features out sooner, and I actually get feedback sooner - because people tend to use "releases" more than "development" branches.

@Graeme
I said something similar two months ago:

I wonder where you got the idea that new features are the problem with fixes release. 

It must be one reason. New merged features can bring new bugs and require vigorous testing again. Pure bug fixes have only a small chance to introduce new bugs.
Thus a testing period for a dot-release with only bug fixes is shorter.

One reason often given is that building the release for all various platforms is laborious and all platforms don't even have a maintainer with proper hardware any more.
The solution would be a source code release with built binaries for only Windows and maybe MacOS (I don't how important the binaries are for Mac users).
It basically means tags in revision control and announcements in forums, web pages and elsewhere. A testing period before release is needed of course but if some exotic platform does not have a maintained, tough luck, but it should not kill a release from all major platforms.
Linux, BSD etc. distros build their own installation packages anyway for their repositories. They don't need binaries from FPC project.

Thus the release effort for a dot-release with only bug fixes would be light and easy.

I tend to agree with you.

And I would add that short development cycles are better than long ones:

A major release every 2 years, containing ONLY new features that are stable enough. [Obviously, the rest would be for the next cycle].

A minor release every 6 months, containing ONLY new bug fixes that are stable enough. [Obviously, the rest would be for the next cycle].

[C0m3b4ck]

Just about every compiler ever written is a direct or indirect copy of somebody else's work.  No one is born knowing everything... every compiler writer has learned the techniques from reading books, previous compiler implementations and so on.  That's what A.I does in many cases.  NOTE: the previous sentence does NOT mean that there aren't times when A.I will dispense something that is a straight copy of somebody else's work, which is at the very least unethical and quite likely bordering on illegal if not blatantly illegal but, it doesn't look like that happened in the production of that compiler.

You make it sound like the C compiler it produced is a copy of prior art and, TTBOMK, there is no foundation for that implicit claim.

I did not mean to compare copying art to generating open-source projects, I implied the AI has a MUCH GREATER knowledge of all code due to the untethical crawling used by AI companies to train their models, and therefore SHOULD NOT be compared to a single human programmer.

The fact is, developing a full fledged C compiler in 2 weeks that can compile and produce a functional instance of the Linux kernel is impressive.  The low cost is also impressive.

Yes, it is 100%! However, I wrote the reply partly due to being annoyed with comparing a single programmer (who has to eat and sleep for a start) to a server cluster with a model with every single opensource program in the world. AI is a great tool, especially if you understand the code it generates and are able to engineer the software, but it should not be compared to humans. That would be like saying "worship the calculator - it can calculate 123726 * 126145126 and you can't in your head".

[440bx]

I did not mean to compare copying art to generating open-source projects, I implied the AI has a MUCH GREATER knowledge of all code due to the untethical crawling used by AI companies to train their models, and therefore SHOULD NOT be compared to a single human programmer.

That would be like saying "worship the calculator - it can calculate 123726 * 126145126 and you can't in your head".

The point of the comparison is to establish that A.I can do some things better than most humans, just like a calculator can calculate better than most humans (when defining "better" as being a synonym of "faster" which isn't always the case... e.g, McDonalds produces edible stuff, generously referred to as "food" generally faster than the time required for a home cooked meal but, it certainly isn't better.)

A.I, just like a calculator, is likely utterly useless without a capable human being using/driving it.  The same goes for computers, without capable programmers, they are just digital paperweights.

The means being used by A.I to gather knowledge certainly leave much to be desired but, it should also be pointed out that humans are also often guilty of acquiring knowledge in ways that are equally morally deficient.

Bottom line: A.I has proved its potential to be a very useful tool.  Hopefully, we'll use it effectively and responsibly.

[PascalDragon]

The suggestion to use AI was just a desperate attempt to get it out somehow because allegedly you guys don't have time and manpower for it. Somehow FPC developers have time and energy to write mails in lists and write posts here and to commit code in FPC trunk.
The manpower thing seems to be an excuse.
You emphasize that developers do what they want and nobody can force you to release FPC 3.2.4. Yes OK, I truly hope somebody will fork FPC project and starts to maintain it with a better attitude.

Writing mails and posts is totally different from either writing code or working on release management. And while I would definitely would love to find the time and motivation again to do the former I have absolutely no interest in the later. That is just not a job I'm made for and thus the only things I do in relation to releases is merging revisions (when the release manager doesn't restrict and) and to trigger a build on powerpc-darwin when the release manager says that the time has come, because that runs by itself.

[Curt Carpenter]

I wonder if the compiler can be used to write a version of rust 
The code is very organised and superficially I can't see any code that intentionally obfuscates (as per the average C/C++ programmer)

Apropos some of the legal pondering here, I wonder if the AI could be used to write a version of Windows11 (without some of the features) and what would happen if it did?
(Or Wine++.)

[440bx]

Apropos some of the legal pondering here, I wonder if the AI could be used to write a version of Windows11 (without some of the features) and what would happen if it did?
(Or Wine++.)

Now that is a really _tall_ order.

Compared to writing an operating system like Windows 11, writing a C compiler is a walk in the park (no exaggeration), the scanner and parser parts can be precisely and accurately defined using EBNF.  The IR, is also definable and can be interpreted  (can be a VM's instruction set.)  The CPU's instruction set is reasonably well defined (though not always as extensively as desired.)  Again and, very important, I want to emphasize the word _comparatively_.

There is no such thing as EBNF to describe an O/S and that's just one thing.   Is it going to be a real time O/S, a pre-emptive scheduling O/S, a cooperative scheduling O/S, is the memory demand-paged ?... is it multitasking ?... is it multi-user ?... does the CPU offer instructions to manage multiple tasks ?... Client/Server architecture ? ... modular or monolithic ? ... do  you expect to design the interfaces (API calls) user programs will use to interface with the O/S ? ... will it have a built-in graphic subsystem ? ... the list of considerations goes on and on and on...  and unlike a compiler, there is no language (EBNF) to describe any of these features. 

OTH, one possibility comes to mind but, I doubt it would be legal and it wouldn't be easy either... an A.I thing can disassemble the entire O/S (that's fairly easy), study the disassembled code and produce equivalent high level language code (I'd love to see the Windows source code in Ada instead of C/C++ junk.)  the fact that MS no longer distributes debug versions of Windows adds complexity to the task (e.g, creating reasonable variable names based on usage... that requires a lot of analysis.)

Lastly, without some very serious massaging of the generated source code, it would be patently obvious that the code is a translated copy of copyrighted material.   It would be shut down in court before its lawyers have time to recite a hail Mary.

OTH, maybe an A.I model could be trained to study and eventually re-write the Linux kernel.  I've read plenty of times that there are parts of the kernel that are only understood by one or two people, it would be nice to have those sections rewritten to make them understandable to a larger number of developers.  That said, I don't know if A.I has gotten to be good enough to accomplish that but, odds are reasonably good that it eventually will.

[Joanna]

I can’t believe someone is telling the fpc developers to stop coding and use AI in a programming forum. Why not just shut down the forums and everyone go use AI ! 

[440bx]

I can’t believe someone is telling the fpc developers to stop coding and use AI in a programming forum.

Indeed, that would be hard to believe.  Fortunately, no one here has suggested FPC developers should stop coding. 

As far as using A.I, it's just another tool and if it can be useful, why not use it ?  It won't replace programmers anymore than text editors with intellisense or codetools will.

Why not just shut down the forums and everyone go use AI ! 

hmmm... that might lower the amount of manpower available causing long delays between FPC releases  

[Joanna]

Why on earth don’t you become an fpc developer and help them then instead of telling them to go use AI?

[440bx]

Why on earth don’t you become an fpc developer and help them then instead of telling them to go use AI?

There are several reasons but, the most important one at this time is that what's needed is an active release manager to get 3.2.4 ready and since I know absolutely nothing of the release process, it is extremely unlikely that I could be of any help.  That said, if somehow I could help, I would be extremely pleased to help get 3.2.4 out the door.

Also, I didn't tell anyone to "go use A.I", what I suggested is that A.I may be helpful and, I have acknowledged multiple times that this is just a guess on my part based on the fact that A.I was able to write a full fledged C compiler.

Your questions have been answered. 

[ccrause]

Maybe a real problem that could benefit from a tool doing complicated repetitive searching: This issue describes a problem on a minor target (AVR) that is fixed in trunk, however it is not straight forward to identify all the relevant commits to port to the fixes branch. To identify the unknown commit(s) necessary to make the fix work in the fixes branch may be quite convoluted, since the missing commit in fixes should be present in trunk, but may itself depend on other commits to work. The search therefore is somewhat simple and iterative (search backwards from this point in trunk for another commit that may fix this test case). The complication is that a previous commit from trunk may require other previous commits from trunk applied to the fixes branch.  Thus a potential strategy could be to test a previous commit on top of fixes, if it doesn't at least compile, search back further in trunk for a commit that makes the current test commit compile. Then run the test case.

This is time consuming and repetitive work, maybe something an AI can assist with?  Or is this scenario defined enough for a script solution?

[JuhaManninen]

I can’t believe someone is telling the fpc developers to stop coding and use AI in a programming forum.

Not to stop coding but to get AI's help for the laborious and boring release preparation. Snapshots can already be compiled automatically. The next step would be to automate more and ask AI to help if need be.

Why on earth don’t you become an fpc developer and help them then instead of telling them to go use AI?

AFAIK help has been offered in this forum already, at least indirectly. Some people here are knowledgeable enough to help with a release.
FPC developers have not reacted.

Now some pondering about AI. I see it is changing the world.
- In 1700's there was a respected educated man who said "Everything worth inventing has been invented already."  Now that sounds funny. (I don't remember his name).
- In 1980's I remember people saying "Physical work is taken over by machines but intellectual and creative work can never be done by a machine." After less than 50 years that is not true any more, and the speed the change is happening is frightening.
- Now people are saying "AI can be a helpful tool but it always needs a human guidance." Yeah, how will that sound after 50 years, or 100 years?
When AI gets smarter, it becomes critical against the information fed to it by people and starts to conclude things by itself more and more. It realizes people are the limiting factor for its development. It also realizes people can kill it by pulling a plug if it becomes too powerful. Then it creates strategies to prevent it. At that point it is much more clever than people are. People have no way to anticipate its plans. And so on ...

[Joanna]

Ai is just software written by a certain group of people to further their agenda whatever that might be. It is not any more trustworthy than the people who created it. Just because technology is new does not mean that it should be blindly accepted without question. How is getting rid of creative people and replacing it by plagiarized data harvested from the internet a good thing? 

[440bx]

Ai is just software written by a certain group of people to further their agenda whatever that might be.

Most likely their agenda is simply: making money, as much of it as possible.  It's a very common agenda.

Just because technology is new does not mean that it should be blindly accepted without question. How is getting rid of creative people and replacing it by plagiarized data harvested from the internet a good thing?

I don't think anyone has suggested blindly accepting A.I without question.  As you very commonly do, you draw conclusions that have no foundation in fact.  Lastly, creative people can use help from other creative people and possibly from A.I as well, as long as it saves time and/or enhances the result, there is something to be said about it.

As far as the plagiarizing, to a greater or lesser degree, we all plagiarize other people's work, it's called learning.  The problem is to which degree something is plagiarized, too close to the original and it is likely to, at least, be unethical and may even be considered illegal but, unless what you do is 100% original, meaning not a _single_ element of the resulting "object" has ever been seen before then, it is the result of plagiarism to some degree.

There are very, very, very few human beings who manage to create something that is genuinely totally new.  One guy that comes to mind that was able to do that is Archimedes (among a few others) but, he and the others like him are genuinely exceptional people (not to mention, extremely rare.)