I was testing out ORT on a couple of projects and noticed that it was rather slow to analyse a poetry based python project (https://github.com/netomi/otterdog). The project has a lock file so I was assuming that the dependency resolution should be rather fast.
After some debugging, it turns out that ORT calls python-inspector on the requirements file that is exported from poetry:
13:42:35.163 [DefaultDispatcher-worker-1] INFO org.ossreviewtoolkit.utils.common.ProcessCapture - Running 'python-inspector --python-version 311 --operating-system linux --json-pdt /tmp/ort-PythonInspector5612839292643848503/python-inspector5965944598659711246.json --analyze-setup-py-insecurely --requirement /tmp/ort-Poetry1934851817638935981/requirements.txt16944187604928218322.tmp --verbose' in '/tmp/ort-Poetry1934851817638935981'...
13:43:10.194 [DefaultDispatcher-worker-1] INFO org.ossreviewtoolkit.plugins.packagemanagers.python.Poetry - Generating 'requirements.txt7902172682984572843.tmp' file in '/home/tn/workspace/eclipse/otterdog' directory...
but as you can see from the log, it takes a while for python-inspector to resolve the dependencies although they are all pinned (from the lock file).
Digging into the code of python-inspector, I figured various performance improvements that could relatively easily be applied. I created a PR at aboutcode-org/python-inspector#163 .
Running this version of python-inspector with ORT on the same project, leads to far better results (see timestamps):
13:45:06.460 [DefaultDispatcher-worker-1] INFO org.ossreviewtoolkit.utils.common.ProcessCapture - Running 'python-inspector --python-version 311 --operating-system linux --json-pdt /tmp/ort-PythonInspector13905845346354005296/python-inspector9105449330839367212.json --analyze-setup-py-insecurely --requirement /tmp/ort-Poetry17813648676616128042/requirements.txt17803415363049372639.tmp --verbose' in '/tmp/ort-Poetry17813648676616128042'...
13:45:17.402 [DefaultDispatcher-worker-1] INFO org.ossreviewtoolkit.plugins.packagemanagers.python.Poetry - Generating 'requirements.txt13258899092414060651.tmp' file in '/home/tn/workspace/eclipse/otterdog' directory...
with the exact same output. When you have multiple scopes defined in your project (in my case I have 5), this improvement can really sum up, I could bring the analysis from 2min down to 30s.
I would be happy if there is some feedback on the PR so we can get that into ORT asap, as I am currently investigating the ability of ORT running license checks automatically via GitHub actions and everything that speeds up the analysis is greatly appreciated.
I was testing out ORT on a couple of projects and noticed that it was rather slow to analyse a poetry based python project (https://github.com/netomi/otterdog). The project has a lock file so I was assuming that the dependency resolution should be rather fast.
After some debugging, it turns out that ORT calls python-inspector on the requirements file that is exported from poetry:
but as you can see from the log, it takes a while for python-inspector to resolve the dependencies although they are all pinned (from the lock file).
Digging into the code of python-inspector, I figured various performance improvements that could relatively easily be applied. I created a PR at aboutcode-org/python-inspector#163 .
Running this version of python-inspector with ORT on the same project, leads to far better results (see timestamps):
with the exact same output. When you have multiple scopes defined in your project (in my case I have 5), this improvement can really sum up, I could bring the analysis from 2min down to 30s.
I would be happy if there is some feedback on the PR so we can get that into ORT asap, as I am currently investigating the ability of ORT running license checks automatically via GitHub actions and everything that speeds up the analysis is greatly appreciated.