[Tts-project] Versioning of mbrola-LANG packages (Was: Maintaining mbrola-LANG packages in tts-team team?)

[Samuel Thibault]

Hello,

Andreas Tille, le mer. 25 févr. 2026 09:31:53 +0100, a ecrit:
> I realised while the actual data file is the same in alls packages I
> checked the metadata are not.  For instance we have
> 
> --- old/mbrola-voices/p/mbrola-ar1/README.txt  2018-04-18 21:25:32.000000000 +0200
> +++ new/mbrola-voices/p/mbrola-ar1/README.txt  2026-02-24 21:34:51.000000000 +0100
> @@ -119,7 +119,7 @@
>  If  you  have  not copied   the MBROLA software   yet,  please consult
>  the MBROLA project homepage and get it:
>  
> -        http://tcts.fpms.ac.be/synthesis
> +        https://github.com/numediart/MBROLA/
>  
>  Copy ar1.zip into your target directory and unzip it : 
>  
> @@ -168,5 +168,5 @@
>  31, bvd Dolez, B-7000 Mons, Belgium
>  tel: +32 65 374133  fax:374129            
>  email : mbrola at tcts.fpms.ac.be
> -web   : http://tcts.fpms.ac.be/synthesis
> +web   : https://github.com/numediart/MBROLA/
>  --------------------------------------------------------------

Some voices may have not been updated indeed. The tcts website has been
shut down for several years now, the github repository should be used
now.

> Now comes my question:  The UDD query
> 
>   udd=> select distinct source, version from sources where source like 'mbrola-%' and release = 'sid' order by version;
> 
> uncovers quite a number of different versioning schemes.  The
> get-orig-source script intended to keep the leading version number as
> long as it is no date string.  I wonder where such versions might
> come from.

They normally come from the respective README files. Either an explicit
version, or an explicit date, used as such, or nothing explicit,
in which case I used 0.0.date to avoid conflicting with any scheme
that would be introduced later, where the date is the date that was
originally recorded in the upstream zip file (lost while migrating to
github, but still available in the earliest debian uploads).

> If there is no official version numbering I think it makes sense to
> settle with either
> 
>    20200331.fe05a0c-1
> 
> or even only
> 
>    20200331-1
> 
> (since chances for multiple commits on one day are pretty low) for *all*
> mbrola-LANG packages.  This version would be informative for our users
> and IMHO reflects the reality better than the current version numbers we
> have.  What do you think?  If you agree, what version scheme (including
> commit ID or without it) would you prefer?

I'm not sure. While the upstream voices packages did mostly have version
numbers, AFAIK they have never been updated since then, so most probably
never will (the people there probably even don't have the source
recordings any more). The https://github.com/numediart/MBROLA-voices/
repository was built to make sure that the mbrola voices are not lost,
and it has indeed not seen any update since then.

But taking 20200331 as date would provide much less information, it
could even lead users into thinking that the voices themselves have seen
an update, which is not true ; while the current version numbers are at
least related to original upstream information.

Also, changing the numbering scheme would mean re-uploading the .orig
files, which are quite large, for questionable value.

Samuel