Cache the feature scan results #503
Conversation
|
This is a very valuable contribution - a much-needed optimization. Many thanks! The build fails, sadly ... could you try to run the build.bat batch file from the command line on your development machine and see if that runs through? |
|
Thanks for writing and maintaining Pickles, it's a great tool. This PR is a modest contribution to improve performance on large solutions. This one brings a 3-fold improvement wth no logic changes. If it turns out to be an appropriate contribution, I have a few other changes to avoid re-creating some objects to bring some more improvement. "./build" works fine on my machine. The Appveyor failures is strange ("SocketException (0x80004005): An existing connection was forcibly closed by the remote host"), so I've made a dummy change to retry it. |
|
The build succeeded 😄 |
|
Great, looks simple and if the unit tests pass I trust that it does its job :-) Could you make an entry in the change log? In changelog.md under Unreleased -> Fixed. Update your repo first because I merged another PR that also contains a change to the changelog.md |
|
I will welcome any and all PRs you may send my way! |
|
I've updated the Changelog. Build is still green. |
On a solution with about 6000 test scenarios, Pickles.exe takes about 16 seconds on my machine to process. This is with the following options: Pickles.exe -f xxx -o xxx --trfmt=nunit3 --enableComments=false -df=json -lr=xxx Profiling shows that a large part of this time is spent scanning the NUnit results. More precisely, scanning by feature name is done repeatedly on the xml results. In this commit, the time is reduced to 5 seconds by scanning the xml once and storing the results in a C# lookup, which is optimized for fast-access by key. Unit tests pass. The resulting json files are identical (except "GeneratedOn" datetime, obviously).
|
Thanks for your contribution! |
|
Thanks 😄 |
|
Released in version 2.18.0 |
On a solution with about 6000 test scenarios, Pickles.exe takes about 16 seconds on my machine to process.
This is with the following options:
Pickles.exe -f xxx -o xxx --trfmt=nunit3 --enableComments=false -df=json -lr=xxx
Profiling shows that a large part of this time is spent scanning the NUnit results. More precisely, scanning by feature name is done repeatedly on the xml results.
In this commit, the time is reduced to 5 seconds by scanning the xml once and storing the results in a C# lookup, which is optimized for fast-access by key.
Unit tests pass.
The resulting json files are identical (except "GeneratedOn" datetime, obviously).