r358062(ncurses) breaks installed ports, howto check?

Thu Feb 27 20:35:33 UTC 2020

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

Am Tue, 25 Feb 2020 05:02:54 -0500
Thomas Dickey <dickey at his.com> schrieb:

> On Tue, Feb 25, 2020 at 10:27:37AM +0100, Baptiste Daroussin wrote:
> > On Tue, Feb 25, 2020 at 04:19:56AM -0500, Thomas Dickey wrote:  
> > > On Mon, Feb 24, 2020 at 08:28:03PM -0500, Thomas Dickey wrote:  
> > > > On Mon, Feb 24, 2020 at 06:35:16PM -0500, Thomas Dickey wrote:  
> > > > > On Mon, Feb 24, 2020 at 06:25:30PM -0500, Thomas Dickey wrote:  
> > > > > > On Tue, Feb 25, 2020 at 04:37:11AM +0900, Yasuhiro KIMURA wrote:  
> > > > > > > From: "O. Hartmann" <ohartmann at walstatt.org>
> > > > > > > Subject: r358062(ncurses) breaks installed ports, howto check?
> > > > > > > Date: Mon, 24 Feb 2020 20:19:59 +0100
> > > > > > >   
> > > > > > > > After r358062, many installed ports do not work anymore on several running
> > > > > > > > systems (CURRENT). /usr/src/UPDATING states one should reinstall all ncurses
> > > > > > > > depending ports, but no hint is given! Can someone mitigate this lack of
> > > > > > > > information? Is there a simple way to check what ports installed on a system
> > > > > > > > rely on ncurses provided by the system?  
> > > > > > > 
> > > > > > > Check thread starting with following message.
> > > > > > > 
> > > > > > > https://lists.freebsd.org/pipermail/freebsd-ports/2020-February/117710.html  
> > > > > > 
> > > > > > That's a start, but it gives an overly-broad approach, saying that
> > > > > > anything linked to the ncurses library has to be recompiled.
> > > > > > 
> > > > > > The ABI change is just to the (upper-level) curses interface.
> > > > > > Most of the programs you'll have in ports use the (lower-level) termcap
> > > > > > or terminfo interfaces.
> > > > > > 
> > > > > > For example gettext uses terminfo (not curses).
> > > > > > 
> > > > > > Curses applications use initscr or newterm (nm helps).
> > > > > > I have a script which uses nm to tell me which interface is used.
> > > > > > 
> > > > > > Actually, in my own ports, I don't see any which would be affected,
> > > > > > since all of the curses applications are the utilities for ncurses
> > > > > > (or for my testing of ncurses).
> > > > > > 
> > > > > > Here's an example of what it tells me
> > > > > > (n5==ncurses5, tc=termcap, ti=terminfo):
> > > > > > 
> > > > > > ti	bison
> > > > > > n5*+ti	captoinfo
> > > > > > n5*+ti	captoinfo6
> > > > > > n5*+ti	clear
> > > > > > n5*+ti	clear6
> > > > > > n5+tc	ded
> > > > > > n5+ti	dialog4ports  
> > > > > 
> > > > > actually this one isn't one of mine (needs to be recompiled)
> > > > > 
> > > > > But for the rest - recompiling would be a waste of time.  
> > > > 
> > > > ...that's just looking at /usr/local/bin.  I see Millard's list
> > > > includes /usr/local/lib.  I have some of those:
> > > > 
> > > > c3+tc	libXvMCr600.so
> > > > tc	libedit.so
> > > > tc	libedit.so.0
> > > > ti	libgettextsrc.so
> > > > tc	libreadline.so
> > > > tc	libslang.so
> > > > ti	libtextstyle.so
> > > > c3+tc   libvulkan_radeon.so
> > > > 
> > > > that is, mesa-dri uses curses, but libedit and libreadline do not.
> > > > 
> > > > I have llvm80, but that doesn't live in either of /usr/local/bin
> > > > or /usr/local/lib.  It's in its own directory (with a script in
> > > > the former pointing there).  It uses curses (and is not a quick
> > > > recompile).  
> > > 
> > > now... I discussed this briefly with the developer on the 20th of January.
> > > 
> > > ncurses can be configured/built to use the old binary interface (ABI 5),
> > > but he ran into some problem doing that, and decided to use ABI 6.
> > > 
> > > At the time, I had supposed that cost would be contained by the system
> > > builders, not considering the impact on users rebuilding ports.
> > > 
> > > Depending on where FreeBSD is along that path, it might be worth
> > > revisiting, to see exactly why the build with ABI 5 did not work.
> > >   
> > I think we should anyway move on on ABI 6, which is why I didn't insist on ABI 5
> > when I could not find the root cause of it not working. In the futur we will
> > update ncurses on more regular basis, and ABI 6 is what everyone else is using
> > so probably what is the most tested.
> > 
> > if I understand correctly upstream build system, staying on ABI 5 disable many
> > features that are available in ABI 6.  
> 
> yes... the wide-character configuration in ABI 6 adds a lot of stuff.
> 
> ABI 5 is for compatibility; narrow-character applications (and termcap)
> wouldn't benefit as much from ABI 6.
>  
> > For those interested my current plan with ncurses is the following for freebsd
> > 13.0:
> > - split terminfo from ncurses (like the upstream build system) so a futur change
> >   in high level ABI will impact less consumers  
> 
> offhand, I see that 2/3 to 3/4 of the consumers use termcap or terminfo.
> 
> My script tells me that - I've not integrated that into something that
> would relate the results to packages, since I was mainly interested in
> comparing curses/terminfo/termcap usage on a variety of systems.
> 
> (I recall there's some problem using it with Solaris - when I revisit
> that and get "enough" data, I'll probably make a nice table on a webpage).
> 
> > - only keep the widechar version of ncurses in base
> > - add the pkgconf file in base to ease the ports tree.
> > 
> > Again for people who don't want to rebuild everything compat12x package has been
> > created and can be used to have all binaries working.
> > 
> > Best regards,
> > Bapt  
> 
> 
> 
Thank you all very much for the fast response.

I started sorting out the worts problems via "pkg check -B" which gave me some valuable hints.
I decided to recompile all ports on those machines where we compile traditionally all ports.

Kind regards,
oh

- -- 
O. Hartmann

Ich widerspreche der Nutzung oder Übermittlung meiner Daten für
Werbezwecke oder für die Markt- oder Meinungsforschung (§ 28 Abs. 4 BDSG).
-----BEGIN PGP SIGNATURE-----

iHUEARYIAB0WIQSy8IBxAPDkqVBaTJ44N1ZZPba5RwUCXlgoBQAKCRA4N1ZZPba5
Rx1LAQDwLuWTzNSLVAUhnZbpuzUxysTO/U7+vucSl5fiM5Ou3wD/RgkCbTZ699M+
g26HzGXU8v3yz2P043I7Gwi17JshmQY=
=4A9a
-----END PGP SIGNATURE-----