On Sun, 25 Oct 2026 at 19:27, Daniel Sahlberg <[email protected]>
wrote:

> Den tors 22 okt. 2026 kl 00:58 skrev Pavel Lyalyakin <[email protected]
> >:
> >
> > Hello,
> >
> > While working on INSTALL, I compared the minimum dependency versions it
> lists with the checks in the build systems. It appears that the checks are
> inconsistent across the build systems and that some of the minimum versions
> are IMHO too old to even be mentioned or checked for.
> >
> > So let's consider dropping minimum versions from INSTALL entirely and
> rely on the build system checks.
> >
> > It seems that mentioning minimum versions in INSTALL is a practice from
> an early era of SVN development, when it made sense and actually mattered.
> The guide said exactly which versions had to be used to build SVN, and
> these versions were recent at that time. BTW, some of the min versions,
> like BDB 4.0.14, are exactly from that era.
> >
> > The situation today is quite different. Most of the minimum versions are
> 10 to 25 years old: BDB 4.0.14 was released in 2001, autoconf 2.59 in 2003,
> httpd 2.2 in 2005 (and it has been EOL since 2017), APR-Util 1.3 in 2008
> and APR 1.4 in 2010. So I believe that SVN has very modest minimum
> requirements, and current Linux distributions, such as RHEL 8, meet all of
> them out of the box. Even RHEL 7 (extended support until 2029) meets all of
> the except SQLite 3.24.0 and CMake 3.20 that are new in SVN 1.15.0.
> >
> > INSTALL mostly duplicates the checks that are already present in the
> build systems and requires additional work to be kept in sync by hand. I've
> updated some of the minimum versions in INSTALL recently, but I think that
> it would be best if the build systems were the single source of truth.
> Hence, I'd drop the min version numbers from INSTALL.
>
> LGTM. Having multiple places to maintain only mean we will forget to
> update some.
>
> +1.

[...]

-- 
Ivan Zhakov

Reply via email to