On 03/19/2016 12:07 AM, Ken Moffat wrote:
On Fri, Mar 18, 2016 at 11:14:05PM +0100, ALZ (phyglos.org) wrote:
On 03/17/2016 11:38 PM, Ken Moffat wrote:
On Thu, Mar 17, 2016 at 11:07:34PM +0100, ALZ (phyglos.org) wrote:
Has anybody found something like this? I don't recall any "make -j1 install"
needed in any other package.
I've never used more than -j1 when installing, I regard it as not
worth the risk. Similarly, for tests (unless somebody has noted
that they work ok like that).
There is/was a ticket raised against openssl for 7.10, noting that
parallel install sometimes failed.
ĸen
Yes, doing "-j1" for all packages is a good choice, as it is running all
available tests before a "release" build. Long build, but no doubts about
how it was build. (I think LFS SBUs are also calculated for -j1)
To be clear: I use -j1 for tests (if I run them) unless advised that
running in parallel is ok and useful (for gcc the tests run as
normal, but the results are all jumbled up and the contrib file that
used to sort out the results no longer seems to do anything) and for
install.
I see your point and I think I'll take some hint to test in my build
process.
But that's the opposite the book (it seems to me) does. By advising in
specific spots when parallel "make" or "make chech" are known to fail,
one can infer that setting MAKEFLAGS "globally" is safe along the rest
of the instructions.
For everything else in a normal build I normally use -JN [ for
values of N between 3 and 8, depending on what else I am doing ]
when I am building the package.
For the book, most of the time I use -j1 to get an accurate
measurement (based on how long it took the new system to repeat an
SBU : usually, slower with each new toolchain, and not consistent to
more than about 5 seconds even for the same toolchain) on an SSD.
At the moment, one of the desktop packages for which I have a ticket
(shared-mime-info-1.6) builds ok with -j8 on my haswell (I very much
doubt that more than 2 or 3 jobs actually run in parallel) but can
only build if I force -j1 on my SandyBridge i3 and my Kaveri A10.
I can't recall which one, but some other "essential" package from FLS
(not BLFS) does not build for me in parallel. I assumed it was a local
issue and haven't asked about that and moved on.
In the book there are several "make -j1" for compiling or "make -j1 check"
for testing clearly marked. However I haven't found any "make -j1 install"
yet. This is what I found unusual in this package.
As I said, I only ever use -j1 for installs (unless a package decides
to override that: not everything obeys the user's CFLAGS, that does also
apply to MAKEFLAGS in at least one weird package which is not in the
book [ MediaInfo ] - I hack it to use what I tell it, because on the
i3 (again) it brought the box to its knees running too many jobs in
the build).
When I put my hat of LFS builder I try to stick as much as possible to
the book. Whenever I depart I accept the price to pay. I'm now having
trouble in compiling openssl with ccache enabled and I don't expect the
book or the editors to solve that.
Anyways, looks like both Arch as well as Gentoo have chosen to build this
package forcing -j1 and ignoring any user defined MAKEFLAGS. They probably
had this issue before.
Thanks!
Alz.
Sounds likely - I don't think many people have any need for it.
/me goes off to look at my scripts: I've been using -j1 to build
s-lang since last September, I presume I needed that when I was
testing gnome for 7.8 on my i3.
So it seems to me that the question regarding S-Lang is solved. And I've
taken some ideas to test in my build process. That's good business. ;-)
Thank you.
Alz.
--
http://lists.linuxfromscratch.org/listinfo/blfs-dev
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page
