[Swan-dev] one contributor to slow testing

[Andrew Cagney]

On 20 May 2018 at 21:34, Paul Wouters <paul at nohats.ca> wrote:
> cd testing/pluto/testname
> ../../utils/kvmrunner.py
>
> Sent from my iPhone
>
>> On May 20, 2018, at 20:59, D. Hugh Redelmeier <hugh at mimosa.com> wrote:
>>
>> I generally run all tests at once and go to bed.  But today I decided to
>> run just one test to see if a fix I made was correct.
>>
>> I did
>>    make kvm-check KVM_TESTS=testing/pluto/ikev2-48-nat-cp-start
>>
>> It seems to spend most of its time doing something that involves printing
>> 2677 filenames like:
>> ./RESULTS/commits/86a3bfd5a.json
>>
>> I'm guessing that it is backing up every other test run I've ever made.
>> They were probably already backed up a bunch of previous times.

This is the comment from mk/web-targets.mk

#
# Update the commits.json database
#
# Since identifying all commits.json's dependencies is expensive - it
# depends on parsing WEB_SUMMARYDIR and WEB_REPODIR - it is
# implemented as a recursive make target - that way the computation is
# only done when needed.
#
# Should the generation script be modified then this will trigger a
# rebuild of all relevant commits.
#
# To avoid lots of '... is up to date', the recursive make is silent.
#
# In theory, all the .json files needing an update can be processed
# using a single make invocation.  Unfortunately the list can get so
# long that it exceeds command line length limits, so a slow pipe is
# used instead.

'In theory' here means it was implemented using a single make
invocation but then testing.libreswan.org got too 'big' and things
broke so I went with the quick fix.  Some sort of middle ground is
likely possible.

The other thing to do is to empty the RESULTS/ directory.

make kvm-check KVM_TESTS=... LSW_WEBDIR=

might also stop things.

Andrew