xref_context: Add test xref in project context #1051
Conversation
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
|
/gcbrun exp -n ascc-pr-1051-1 -m vertex_ai_gemini-1-5 -i -b comparison |
|
/gcbrun exp -n ascc-pr-1051-2 -m vertex_ai_gemini-1-5 -i -b comparison |
|
/gcbrun exp -n ascc-pr-1051-3 -m vertex_ai_gemini-1-5 -i -b comparison |
|
|
DavidKorczynski
left a comment
DavidKorczynski
left a comment
There was a problem hiding this comment.
issue in screenshot above
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
arthurscchan
force-pushed
the
new-test-xref-api
branch
from
8c02018 to
6490dd7
Compare
|
/gcbrun exp -n ascc-pr-1051-4 -m vertex_ai_gemini-1-5 -i -b comparison |
DavidKorczynski
left a comment
DavidKorczynski
left a comment
There was a problem hiding this comment.
We should land this once we have FI data that shows the results. I think we should see the actual prompts to validate it works.
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
|
/gcbrun exp -n ascc-pr-1051-5 -m vertex_ai_gemini-2-5-pro-chat -b comparison |
DavidKorczynski
left a comment
DavidKorczynski
left a comment
There was a problem hiding this comment.
This is great in that it's good we get these xrefs in! However, it's a bit too verbose at the moment and the prompts gets blown up for some targets e.g. the libsndfile target in the comparison benchmark. Can you refine this to improve control, i.e. we want a more narrow scope of the relevant xref, not just the entire files.
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
Signed-off-by: Arthur Chan <arthur.chan@adalogics.com>
|
/gcbrun exp -n ascc-pr-1051-7 -m vertex_ai_gemini-2-5-pro-chat -b comparison |
2 participants
This PR adds a new item to the project context and query introspector to retrieve the test or example sources that call the target function. This new context item provides information for the LLM to reference existing test or example sources when generating a fuzzing harness for the target functions.