mailing-list for TeXmacs Users

[Henri Lesourd <address@hidden>]

Wolfgang Jansen wrote:

> Henri Lesourd wrote:
>
> > Wolfgang Jansen wrote:
> >
> > > is a the impossibility of installing TeXmacs because
> > > of errors in C++ code an important bug?
> > These kinds of bug are as much bugs in the GNU toolchain
> > as these are TeXmacs's C++ bugs.
> Are you severe ?

Undoubtedly, there is an epidemic of crass speech, these
days ;-(...


>   The bug is not somewhere in the GNU toolchain,
> it is in TeXmacs: the error message is produced by g++ when compiling
> one of the TeXmacs sources,
Thank you very much, I know that *across versions of gcc*,
compiling the TeXmacs sources can become broken.

Sometimes this kind of thing is because gcc is becoming
more standard compliant, and thus our code is wrong. And
some other times a new version of gcc introduces a more
or less subtle bug, for which we need to implement a
workaround.


> > > Half year ago I tried to install TeXmacs on SUN from sources
> > > (the existing binary is rather old). Installation failed
> > > because of errors in C++ code. My bug report from 12/14/07 seems
> > > have to be forgotten: the last version just downloaded fails as well.
> > > So, I still work with the holy trinity emacs/latex/xdvi.
> > It's hard to debug autoconf-based makefiles, and if we
> > don't have the appropriate machines and systems, it's
> > even pretty impossible.
>
> I don't think that error reports of "configure" are machine specific.

Your view of the GNU toolchain seems exceedingly naive: lots
of things, including the bugs in configure itself, can be
more or less platform-dependent.


> As I mentioned in the bug report the reason is that the installation
> tries to copy/move some files to disk locations not accessible
> by a simple user.

We are well aware of these kinds of issues, although I
did not imagine there were some in the install *TOO* !


> Once again, please have look at the bug report.
> The solution may be just to skip over those parts of "configure"
> after successful generation of "Makefile".

Usually, the solution is not at all easy to implement,
for which directories are world-readable or not is
highly dependent on exactly which UNIX you use.

Thank you for the hint; in any case, posting a mail
about important bugs is always useful, the mailing
list is reviewed more often.


> Sorry, the bug occurs deep in the C++ class dependencies.
> I really do not understand which class to modify to fix the bug,
> and I do not know which dependencies are disturbed
> if I modify something.

If there is a bug in compiling the C++ (not the other
one involving moving files), then it's very probably
related to the version of gcc: currently, the CVS
compiles correctly with gcc >= 4.2.1.


> Even worse, the first C++ error (where installation stops) today
> is at another place (Kernel/Containers/ntuple.hpp), and I do not
> know how many may follow.

A typical C/C++ compiler problem: only the first error
message is significant, usually.


> In any case, I used "g++" for compilation. Is this an exotic compiler?
> How did you manage to avoid the compilation errors?

We use a more recent compiler, very probably. Try:
gcc --version

to obtain the version of yours.


> > Thus: we cannot afford spending very much time
> > on such issues, especially in a context where
> > the problem is also in dealing with debugging
> > the autoconf / GNU toolchain mess, and in
> > circumstances where, as well, users don't
> > want to / cannot help on this.
> The failure of "configure" is annoying but is not the killer.
> The killer is the bug in C++ code.
> Can you really not spend time to eliminate compilation errors?

If we need to be sure it works on *all platforms*,
this is a daunting task.


> > > TeXmacs has lost at least one possible user,
> > > and I warned my students to use such a buggy software.
> > This was not a very sensible idea, for as soon as
> > your students use TeXmacs on a widely spread UNIX
> > brand (e.g. Ubuntu), it works perfectly.
> So, you should restrict your software to specific target platforms only.

De facto, that's more or less what we do.


> The TeXmacs home page says "TeXMacs runs an all major UNIX
> platforms and WINDOWS". You may believe it or not, SUN/Solaris is still
> a major UNIX platform.
The point is that usually, these issues are solved
by the configure + automake toolchain, in such a
way that sooner or later, things become sane again.

But if on a given platform, nobody does the job of
fixing the problems, then we are stuck, for the people
of the core TeXmacs team do not enjoy an access to
all the platforms currently in existence.