From: Beman Dawes (bdawes_at_[hidden])
Date: 2003-09-03 19:16:13
At 10:14 AM 9/3/2003, Matthew Towler wrote:
>Beman Dawes wrote:
> > Are you using run_tests.sh from CVS or your own script?
>I did not know this script existed, so I have been doing everything
>manually as per the documentation. on
>and making the obvious syntactic adjustments for unix
Ouch! My fault. I've added a link (in CVS) to run_tests.sh to reduce
likelihood of it being missed.
>One thing I noticed is that this documentation indicates setting a
>BOOST_ROOT environment variable, whilst run_tests.sh uses boost_root, I
>was under the impression that case mattered - or is this irrelevant as
>boost_root is always passed into the executables on the command line
>rather than automatically referred to?
I don't know the answer to that. Someone familiar with running under Unix
needs to answer.
>I retried the processing of the regression test results using the
>run_tests.sh script and got exactly the same result from
>*** Error: std::runtime_error: boost::filesystem::directory_iterator
>No such file or directory
>BOOST_ROOT (and boost_root in the script) were set to
>so I believe it is the processing program that is adding the spurious
>'0_2' to the path.
I would do several things if that happened here:
(1) run only on a single test (I usually use config_info). That makes
testing simpler and quicker.
(2) inspect the bjam log file to see if anything obvious jumps out.
(3) look at the target directory to see that the expected files are
present. I set -sALL_LOCATE_TARGET=D:\boost-regr, so GCC target ends up
A listing of that directory shows:
08/14/2003 09:08 AM 1,775,667 config_info.exe
08/14/2003 09:08 AM 196,220 config_info.obj
08/14/2003 09:32 AM 4,310 config_info.output
08/14/2003 09:32 AM 4,310 config_info.run
08/14/2003 09:32 AM 51 config_info.test
08/14/2003 09:43 AM 4,771 test_log.xml
The contents of test_log.xml are:
<test-log library="config" test-name="config_info" test-type="run"
<compile result="succeed" timestamp="2003-08-14 13:43:35 UTC"></compile>
<link result="succeed" timestamp="2003-08-14 13:43:35 UTC"></link>
<run result="succeed" timestamp="2003-08-14 13:43:40 UTC">
GNU C++ version 3.3.1
... macro values ...
Where "macro values" is a long list. Is the test_long.xml file present? Are
the various paths correct in it?
If test_log.xml is missing or wrong, the problem is in process_jam_log or
bjam. If test_log.xml looks good, the problem is in compiler_status.
(4) Run compiler_status under a debugger, and trap the call stack at the
point the exception is thrown. By inspecting the call stack, it should be
possible to figure out where the bad path came from. Let me know; we might
be able to add an error check or workaround.
>I also tried the script on IRIX (with the mipspro 126.96.36.199 compiler),
>with a script adjusted to set test_tools and toolset to mipspro. I got
>the following (good) output for the processing
>processing the regression test results for mipspro:
>Usage: bjam [bjam-args] | process_jam_log [locate-root]
> locate-root is the same as the bjam ALL_LOCATE_TARGET
> parameter, if any. Default is boost-root.
>no errors detected
>generating html tables:
>Using /local/buildman/builds/boost-1.30.2/status/bin/config_test.test to
>no errors detected
>However when I look at the generated webpage (cs-IRIX64.html), most of
>the tests are marked as missing. e.g. for regex regex_grep_example_1, 2,
>3, are marked as Missing, whilst regex_grep_example_4 is a Pass.
>Looking at the corresponding regress.log file, all of these four tests
>passed - and when located in $BOOST_ROOT running
>find . -name "*.xml" | grep example_
>so it would appear the output is there - so I do not understand why the
>tests are Missing.
That sounds like compiler_status is being invoked from the wrong working
directory or in some other way is not finding the target directories.
Again, isolate a single test on a single compiler, and follow it through.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk