[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[STINNER Victor]

STINNER Victor added the comment:

> Please go ahead with bininstall-3.patch if you prefer.

Ok done.

> It certainly avoids the problem you reported. Although I still encourage you 
> to not add the dash before the command.

Oops, fixed.

--
resolution:  -> fixed
status: open -> closed

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[Roundup Robot]

Roundup Robot added the comment:

New changeset 94910d210ef4 by Victor Stinner in branch '2.7':
Issue #25696: Don't ignore errors in 'make bininstall' on creating $(LIBPC) 
directory
https://hg.python.org/cpython/rev/94910d210ef4

New changeset d28268c47421 by Victor Stinner in branch '3.5':
Issue #25696: Don't ignore errors in 'make bininstall' on creating $(LIBPC) 
directory
https://hg.python.org/cpython/rev/d28268c47421

--

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[Roundup Robot]

Roundup Robot added the comment:

New changeset c03ef448b5b2 by Victor Stinner in branch '2.7':
Issue #25696: Fix installation of Python on UNIX with make -j9.
https://hg.python.org/cpython/rev/c03ef448b5b2

New changeset 87d96b349ff5 by Victor Stinner in branch '3.5':
Issue #25696: Fix installation of Python on UNIX with make -j9.
https://hg.python.org/cpython/rev/87d96b349ff5

New changeset 8dd594d36704 by Victor Stinner in branch 'default':
(Merge 3.5) Issue #25696: Fix installation of Python on UNIX with make -j9.
https://hg.python.org/cpython/rev/8dd594d36704

--
nosy: +python-dev

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[Martin Panter]

Martin Panter added the comment:

Please go ahead with bininstall-3.patch if you prefer. It certainly avoids the 
problem you reported. Although I still encourage you to not add the dash before 
the command.

--

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[STINNER Victor]

STINNER Victor added the comment:

bininstall-4.patch: IMHO it's overkill and makes Makefile more complex to 
follow. My patch  bininstall.patch proposed something similar, but Arfrever 
asked me to modify bininstall target. IMHO Getting "Creating directory xxx" 
message twice is a minor issue.

It becomes annoying to have nitpicking on my patch, I hate autotools, I hate 
having to care of Makefile, etc. Maybe I should just abandon my change, it 
looks like nobody cares that Python cannot be installed with -j9 anyway :-)

--

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[Martin Panter]

Changes by Martin Panter :


Added file: http://bugs.python.org/file41178/bininstall-4.patch

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[Martin Panter]

Martin Panter added the comment:

I agree your patches should fix your practical bug (I can reproduce it, two out 
of three times). My concern is more about making the code cleaner and less 
likely to grow problems in the future. :) The worse case with patch 3 would be 
the “Creating directory” message being echoed twice. I am just trying to get 
the makefile cleaner.

If that block compiled a file, instead of just creating a directory, then it 
would be a worse problem. The second compile command could truncate the first 
version, causing errors later on.

Also, I don’t believe adding the dash (-) prefix is a good idea here. It tells 
Make to keep going even if the command failed. In this case, if the directory 
already exists, the “if” statement will succeed, and the only way it can fail 
is if $(INSTALL) fails, which I think we would want to know about. The original 
“for” loop under libainstall does not include a dash.

bininstall-4.patch implements my suggestion from before.

--

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[STINNER Victor]

STINNER Victor added the comment:

> $(LIBPC) is a single directory name, so I suggest dropping the “for” loop.

Ok, here is a version without loop. I also added "-" before the if to mimick 
other lines of the bininstall target.

Does it look better now?

> (...) there is technically still a race with “libainstall” and “bininstall” 
> both testing and creating the same directory.

Sorry, I don't understand. Is it an issue in practice to create the same 
directory twice "at the same time"? Are you able to reproduce the bug?

Please try my use case: "./configure && make && make -j10 install" and enjoy 
the partial installation :-) (The python binary is installed in bin/ but it 
doesn't work because the stdlib is completly missing.)

--
Added file: http://bugs.python.org/file41171/bininstall-3.patch

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[Martin Panter]

Martin Panter added the comment:

$(LIBPC) is a single directory name, so I suggest dropping the “for” loop.

Patch 2 should avoid the practical race condition. But there is technically 
still a race with “libainstall” and “bininstall” both testing and creating the 
same directory. Maybe we should factor it out, something like:

$(DESTDIR)$(LIBPC):
@echo "Creating directory $@"
@$(INSTALL) -d -m $(DIRMODE) $@
. . .
bininstall: altbininstall $(DESTDIR)$(LIBPC)
. . .
libainstall: all python-config $(DESTDIR)$(LIBPC)
@for i in $(LIBDIR) $(LIBPL); \
. . .

Looking at the history, the test-then-install code comes from 20 years ago: 
. I would say calling 
install unconditionally (without a test or echo message) might be simpler, but 
that’s probably getting out of scope here.

--
nosy: +martin.panter
stage:  -> patch review

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com

[issue25696] "make -j9 install" fails because bininstall target requires the libainstall target

[STINNER Victor]

Changes by STINNER Victor :


--
title: "MAKEFLAGS=-j9 make install" fails because bininstall target requires 
the libainstall target -> "make -j9 install" fails because bininstall target 
requires the libainstall target

___
Python tracker 

___
___
Python-bugs-list mailing list
Unsubscribe: 
https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com