[asdf-devel] cl-test-grid questions
Dear Anton, I probably do NOT want my cl-test-grid results to be mixed with the usual results, since I want to run with my own version of ASDF which will skew the results. But I DO want to compare my results to an existing baseline. How do I achieve this effect? Where in cl-test-grid is the correct place to hook a call to asdf:upgrade-asdf before trying to load stuff? And to (asdf:initialize-source-registry '(:source-registry :ignore-inherited-configuration)) to make sure I'm testing what's in quicklisp and now my own source checkouts? (or are you doing that part already?) —♯ƒ • François-René ÐVB Rideau •Reflection&Cybernethics• http://fare.tunes.org If a law is unjust, a man is not only right to disobey it, he is obligated to do so. — Thomas Jefferson
06.01.2013, 19:41, "Faré" <fahree@gmail.com>:
And to (asdf:initialize-source-registry '(:source-registry :ignore-inherited-configuration)) to make sure I'm testing what's in quicklisp and now my own source checkouts? (or are you doing that part already?)
I didn't know it is necessary, but added it right now, so it is solved.
I probably do NOT want my cl-test-grid results to be mixed with the usual results, since I want to run with my own version of ASDF which will skew the results. But I DO want to compare my results to an existing baseline. How do I achieve this effect?
Where in cl-test-grid is the correct place to hook a call to asdf:upgrade-asdf before trying to load stuff?
I thought you can have a directory quicklisp-patched/ with quicklisp patched to have newer version of ASDF and also changed (defvar *required-asdf-version* "2.<your version>") in the quicklisp-patched/setup.lisp file, line 26. This will enforce asdf upgrade by quicklisp. Then, if you setup test-grid-agent as described at https://github.com/cl-test-grid/cl-test-grid, instead of calling test-grid-agent:main in your run-agent.lisp script, you do: (setf (test-grid-agent::result-storage-name *agent*) "asdf") (test-grid-agent::test-patched-quicklisp *agent* #p"quicklisp-patched/" "quicklisp 2012-12-23+asdf-patch-01") This will run tests on all the lisp-exe's you configured and upload test results to online storage named "asdf". Without the above customization the results are uploaded into the storage named "main". So your results will not be mixed with usual results. Test results are represented as a simple p-list, so we will be able to extract result from main storage and compare with your results. On the other hand it may be more reliable that you run tests for not-patched quicklisp on your machine too, so that patched and unpatched quicklisp. (test-grid-agent::test-patched-quicklisp *agent* #p"quicklisp/" "quicklisp 2012-12-23") I can assist with all this. We can setup a chat, or skype call to help you to start the tests. After tests are finished, we can contact again and I will show how to access the results and compare them. Just looking ahead, to access the results from online storage the following functions are is used: ;; create local replica (defparameter *r* (test-grid-storage:make-replica "asdf" "db-asdf-snapshot.lisp")) ;; fetch the changes (test-grid-storage:sync *r*) ;; access the data (test-grid-storage:data *r*) Best regards, - Anton
Dear Anton, thanks to your cl-test-grid machine, I could run tests. All the systems that you told me had trouble with the latest ASDF now pass. For the record, I had to: 1- fix ASDF (see commit 2.26.59) 2- patch buildapp (sent a pull request upstream) 3- patch asdf-system-connections (had already sent patch upstream earlier) 4- export LANG=en_US.utf-8 5- use the latest ironclad and nibbles (may or may not have been useful) 6- patch xcvb-bridge (committed upstream, need to release for quicklisp) 7- patch weblocks (sent upstream some time ago) 8- patch qbook (patch sent upstream) 9- patch lisp-executable (patch sent upstream) Exception: * lapack dies with some low-level failure in DGESVD... heap exhausted. I'll blame the build machine's limits. So... things are looking good. I'll wait for everything to be settled before I issue an official release candidate. —♯ƒ • François-René ÐVB Rideau •Reflection&Cybernethics• http://fare.tunes.org Money is like an arm or leg — use it or lose it. — Henry Ford
07.01.2013, 10:29, "Faré" <fahree@gmail.com>:
Dear Anton,
thanks to your cl-test-grid machine, I could run tests. All the systems that you told me had trouble with the latest ASDF now pass. For the record, I had to: 1- fix ASDF (see commit 2.26.59) 2- patch buildapp (sent a pull request upstream) 3- patch asdf-system-connections (had already sent patch upstream earlier) 4- export LANG=en_US.utf-8 5- use the latest ironclad and nibbles (may or may not have been useful) 6- patch xcvb-bridge (committed upstream, need to release for quicklisp) 7- patch weblocks (sent upstream some time ago) 8- patch qbook (patch sent upstream) 9- patch lisp-executable (patch sent upstream)
Exception: * lapack dies with some low-level failure in DGESVD... heap exhausted. I'll blame the build machine's limits.
Yes. SBCL's garbagde collector is conservative is it sometimes uses much more heap than necessary. The most impressive example is f2cl library (a fortran compiler). SBCL can't compile it even with 1GB heap (although these errors are not 100% reproducible, may depend on the system state).
participants (2)
-
Anton Vodonosov -
Faré