[fpc-devel] [RFC] Modernising the FPC Release Process -- Proposal for Review

[Michael Van Canneyt]

On Thu, 16 Apr 2026, Tomas Hajny via fpc-devel wrote:

> On 2026-04-16 11:08, Michael Van Canneyt via fpc-devel wrote:
>> On Thu, 16 Apr 2026, Marco van de Voort via fpc-devel wrote:
>>> Op 16-4-2026 om 00:30 schreef Tomas Hajny via fpc-devel:
>>>> 
>>>> As somebody having been involved in that role in the past (see 
>>>> https://wiki.freepascal.org/Release_Template and e.g. 
>>>> https://wiki.freepascal.org/Release_2.0.4), I can say for sure that the 
>>>> release coordinator / release master / ... certainly doesn't need to know 
>>>> all the details of the compiler, RTL and everything else at the same time 
>>>> :-) - this isn't the issue. From my experience, making sure that people 
>>>> who can tackle specific tasks blocking the release (should it be 
>>>> particular manual steps in the release building process, or fixing some 
>>>> issue which popped out during the build), find needed time for addressing 
>>>> the needed things, is the biggest obstacle.
>>> 
>>> Amen.
>> 
>> 2 things:
>> 
>> It's 2026. The world has changed.
>> 
>> Back in 2022 that was correct reasoning. Today I will ask Claude to fix it.
>> This will seriously reduce the need to wait.
>
> Still, you need somebody able to prompt Claude and review the output - 
> somebody who understands the area which need to be fixed, otherwise you'd 
> apply changes blindly.

I always verify. We can suppose that people who create/maintain a compiler
can judge the solutions the AI put in place.

I also said it 'reduces the need' - that leaves open the possibility that
the AI doesn't know the problem or that the person doing the fix cannot
judge.

No bullet-proof solutions exist.

>
>
>> The risk of this will be smaller if we have a good testing & builder 
>> infrastructure
>> set up: if a non-major platform can be included in that, then the problem 
>> simply
>> does not arise.
>
> I don't think that I've ever opposed that by a single word.

I didn't say that.

>
>> Here at work we exclusively use linux to build Delphi applications for 6 
>> platforms. This is in no way different from what FPC needs to do.
>
> If building for all FPC major targets may be performed on Linux (including 
> Mac OS X and Windows installers) and somebody sets it up that way, I'll be 
> happy. Building for OS/2 cannot be performed that way (at least no directly), 
> but again, that is not an issue and never has been.

I never claimed it was, I'm not pointing any fingers. On the contrary, I
wish to help users having regular releases and at the same time make 
sure the FPC devs can latch on whenever they can.

The regular release train is also supposed to help us prepare: people will
know in advance, so they can prepare. By introducing a freeze period of one
month (ballpark figure) it should give them some time to test, whatnot.

Michael.