[lfs-dev] gcc test failures with current svn

Ken Moffat zarniwhoop at ntlworld.com
Sun Aug 7 13:56:52 PDT 2016

On Sun, Aug 07, 2016 at 03:19:57PM -0500, Bruce Dubbs wrote:
> Ken Moffat wrote:
> > On Sun, Aug 07, 2016 at 02:28:52PM -0500, Bruce Dubbs wrote:
> > > 
> > > Romain, you do know that we run the gcc tests in a limited chroot
> > > environment where gdb is not available, right?  Shouldn't the tests that
> > > need gdb just be skipped?
> > > 
> > Arghh!  I build gdb and strace on my gold builds so that I can debug
> > build failures if any happen in the BLFS packages I need to build
> > before booting.
> > 
> > So, I've got choices - build like the book (boring, but it will
> > determine if I can get through the perl tests) and drop gdb but
> > keep gold : I'll try that if doing it by the book completes and
> > boots.
> > 
> > /me wishes he'd started a build before you updated glibc and
> > binutils, it might have been easier :)
> You might want to start using jhalfs for tests.  I understand you want to
> experiment a bit, but if you want to revert, you can just point it to a
> different, or modified, version of the book.
I might, but it would take time I'm not willing to commit.  I put
/sources on nfs (and so I never build in /sources).  Pierre offered
to consider changes, but I did not pursue the matter - I take the
view that changing the xml instructions should wait until I know what
I want to do.

My scripts are in git, reverting or branching is straightforward
(but limited to only using one branch at a time, even for multiple
machines, because the git tree is in /sources).  I did try building
on multiple machines with local branches (to build different
desktops on different machines), but pushing back the (BLFS-script)
changes was sometimes a bit messy and I haven't done that since

> I admit I have a pretty fast system. but the other day I did a build at -j10
> without any tests and it only took 48 minutes for everything except actually
> building the kernel.  You can also modify the scripts manually for
> individual packages and set breakpoints.

On my i7 haswell, building without tests is, as we say here, a piece
of piss.  But it proves nothing, and the perl hang was in a test.
For my new builds I'm more interested in getting through the
tests, even if (as has become apparent) I don't always read all the
results.  On subsequent builds of the same version, sure, I skip the
tests sometimes.

> I am not suggesting that you change your overall methodology, but perhaps
> add another tool to your toolbox.
> My basic method is:
> Make changes to the book in my sandbox
> Build the book
> # Umount any current virt fs & erase everything on /mnt/lfs except sources
> sh clean-mnt-lfs.sh
> cd ~/jhalfs
> make
> cd /mnt/lfs/jhalfs
> make
>   -- Bruce
I'll think about that, but I loathe editing the XML of either book:
it's something I have to do for commits, but by that stage I usually
know what I want to write.  Sometimes the edit is simple and all I
have to worry about is my phraseology.  Other times, getting it to
render is a PITA, at least in BLFS.  And branching in svn is painful.

I suppose I might look at copying the LFS svn to somewhere else so
that I can use it as a working tree for jhalfs without accidentally
committing it, and writing a script to copy the packages and patches
to /mnt/lfs/sources (I've already got scripts to compare my package
versions and patches to LFS, and to compare my package versions to
BLFS, as well as for checking the other perl modules I build - bash
isn't always the most obvious tool, but it saves me from using php.
;- )  You can alternately read that as "If the only tool you have
is a (choice of) shells, everything looks like a bash problem".

But what I *really* need to do is *use* the damned systems - I
haven't "processed" any of my photo backlog for months, nor used
audacity.  And although my vlcs can now play DVDs with the corrected
kermel config, ISTR that parole and xine were both b0rken for DVDs
on my last build.

Plus, I still need to make proposals/comments on fonts, I also would
like to find time to look at "more interesting" fonts, and a load of
real-life things which I ought to do.  So don't hold your breath!

`I shall take my mountains', said Lu-Tze. `The climate will be good
for them.'     -- Small Gods

More information about the lfs-dev mailing list