JetBrains-Research / JetBrains-Research/TestSpark
Create a prompting strategy for the CUTs whose full definition already exceeds the LLM's token limit
- Dominant language
- Kotlin
- Stars
- 88
- Forks
- 24
- PR merge metrics
- No merged PRs in 30d
Description
# Description
Currently, TestSpark aborts the execution once if the prompt with the least length (i.e., on the class under test definition included) still overflows the LLM's token limit. We may incorporate an additional prompt construction strategy that would include only the method signatures of the class under test.
It will likely degrade the test quality, but at least we generate some tests, thus covering a broader range of classes under test (namely, those with very long implementations).
We can talk about this topic more in some meetings.
# Definition of Done
* [ ] Figure out a prompt construction strategy if the full definition of the CUT is too long to be sent in a prompt alone.
* [ ] Develop and integrate this strategy in TestSpark.
Contributor guide
Assessment
This issue has not been assessed yet.