This is nice idea, but unfortunately does not work for my purpose
of calculatingpass rates for every test suite of regression
testing.
It would fit perfectly if instead of ---dumptest b2 will also have
another option, something like: --dumptestdirs. I mean, the option
which generates the whole list of dirs containing all jamfiles
executed for full regression testing.
Actually, I spent some time trying to generate such list of dirs
based on the output produced by 'b2 -n -a -d0 --dump-tests', but
was not able to find general algorithm
[ wout using specific library names and whout parsing the content
of jamfiles ] which transforms list of tests into the list of
test dirs from which they are launched
[ assuming that each test should be executed only once ]. Here
just couple of examples which demonstarte those difficulties:
1. In 'wave' its 'test' subdir does not contain jamfile like
many other libsdo, but rather wave's jamfile is located inside
internal 'build' subdirectory, and it launches testslocated from
'testlexers' 'testwave' subdirs which are located on the same
level as 'build' itself. When my script sees those wave tests
from --dumptest output, how should it figure out that jamfile
which launches them is located inside 'build' dir?
2. Inside 'unordered' library there are two subdirs with
jamfiles: test/exception and test/unordered. All tests for
'unordered' may be run in two similar ways: either by executing
both those jamfiles, or by executing jamfile which is located on
higher level inside 'test' subdir. But to run every test once
the script should choose one or another way. Now, in a similar
situation for 'core' library, in order to run all its tests one
time my script should run both jamfiles: internal one located
inside core/test, and also another one located in core/test/swap.
How the script should resolve such ambiguity for 'unordered' and
core' libraries?
So, I think that list of test dirs which was present in
status/Jamfile.v2 in boost-1.61.0 was really useful to me. But, of
course if it was created manually, and there is no easy way to do
it automatically, then probably, I should also create it manually
for Boost-1.62 and all later versions.
Sergey
> 2016-09-04 11:45:41 Rene Rivera (grafikrobot_at_[hidden])
> wrote:
> You can get a list of all the tests that get, or will get,
run from b2 with
the following:
> % cd boost-root/status
> % b2 -n -a -d0 --dump-tests
> That will produce, interspersed with other output, lines
like the following:
> boost-test(RUN) "accumulators/count" :
"libs/accumulators/test/count.cpp"
> One of those for each test that is tested. The format is:
> boost-test(<TYPE>) "<LIBRARY>/<TEST>" :
"<SOURCE>"
> Where LIBRARY can be of the form "<library>" or
"<library>/<sublibrary>".
> That should be sufficient to produce the results you want.
And if you use
> just the above list of tests it's likely to be more accurate
than your
> previous method. As the tests runs from the
"boost-root/status" dir are not
> alway the same as those run from the corresponding library
dirs.
> -- >-- Rene Rivera
>>On Sat, Sep 3, 2016 at 5:09 PM, Sergey Sprogis
<sergey.sprogis_at_[hidden]>
>> wrote: