[Swan-dev] proposed patch to re-create version.c every time

[Andrew Cagney]

On Wed, 31 Jul 2019 at 08:56, Antony Antony <antony at phenome.org> wrote:

> Hi
> tangential to "make showversion" discussion, there is an issue that annoys
> me. Now I have a fix for it! If there is no violent opposition against
> this
> proposed patch I would like to apply the attached patch.
>
> version.c dependency is broken while developing and testing continuously
> from
> one directory. Current dependency works great if you always build from a
> clean git tree or run 'make clean' first or build from a tar ball.
> However,
> less ideal for continuous development and testing.
>
>
Right, for the case where the version file really matters, it is correct.
 See the bottom end of:
https://lists.libreswan.org/pipermail/swan-dev/2016-August/001609.html
This is also the least worst option.

For instance, invoking 'make' on an up-to-date build should do 'nothing'.
With this patch applied, repeated makes will instead repeatedly rebuild all
the programs.  In addition to being confusing, and misleading (what did I
change, is a dependency broken) it is also wrong - it cascades as now the
re-built programs are going to needlessly trigger other dependencies (for
instance make check in cavp).

On KVMs, things get worse.  A sequence like (i.e., make certain things are
up-to-date and then run a test):
   $ make kvm-install kvm-test KVM_TESTS=testing/pluto/basic-pluto-01
now has the additional overhead of both running git with the dirty check
and against 9p (which I think we can all agree is expensive, but something
we can't see since it is being hidden under a rock) and then build new
executable.

The problem with git and KVM is also also why I've never bothered to fix
this - any 'technically correct' solution will need to invoke 'git dirty'
within the KVM - is that overhead really worth it - not for me?

Andrew

Here is the case where it breaks.
> You run "make base install-base" starting with a clean git tree.
> Now make changes to the code and commit those changes.
> then "make base install-base". Now the  version.c would be stale.
> ie. pluto --version would not change even though the code changed, and
> commit id, or "make showversion" changed.
>
> ./OBJ.linux.x86_64/programs/pluto/pluto --version
> Libreswan v3.28-526-g5c9063def5-master-s2 XFRM(netkey) esp-hw-offload FORK
> PTHREAD_SETSCHEDPRIO NSS (IPsec profile) DNSSEC FIPS_CHECK LABELED_IPSEC
> SECCOMP LIBCAP_NG LINUX_AUDIT XAUTH_PAM NETWORKMANAGER CURL(non-NSS)
>
> With this proposed patch version.c will be always (re)created.
> "make base' or make programs will have the correct version string.
> It would help me a lot. Are there any side effects if version.c is
> (re)created every time?
>
> Currently I use a work around, "make clean". it has two disadvantages.
> It would add another 30 - 180 sec delay for when you make a change
> re-compile.  Often I forget that and have back track double check.
>
> Let me know what you think. If there are no major objections to this patch
> I
> will apply it, if no comment in week I will apply it:)
> See the attached patch.
>
> -antony
> _______________________________________________
> Swan-dev mailing list
> Swan-dev at lists.libreswan.org
> https://lists.libreswan.org/mailman/listinfo/swan-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.libreswan.org/pipermail/swan-dev/attachments/20190803/c6313b1c/attachment.html>