5 ms·

Human-competitive Patches in Automatic Program Repair with Repairnator

40 points by Dotnaught 8y ago

dang 8y ago
Url changed from https://www.theregister.co.uk/2018/10/17/luc_escape_bug_fixer/ https://www.theregister.co.uk/2018/10/17/luc_escape_bug_fixe..., which points to this.
jboggan 8y ago
I spent last year attempting something very similar in Python (www.duckly.com - unlaunched) and am very impressed. It's an extremely difficult problem and the CI data angle is inspired.
arayh 8y ago
I wonder if the Repairnator is in charge of writing its own commit comments (which I think is technically possible). The simple NullPointerException fixes don't look very impressive, although understandably still quite awesome for a bot. On the other hand, its commit on the eclipse/ditto repo was pretty interesting: https://github.com/eclipse/ditto/pull/151/commits/7ab96e373ecc55e1b942bfff4c9550c8ba5e584d https://github.com/eclipse/ditto/pull/151/commits/7ab96e373e... I'm quite surprised how it determined the usage of a function it did not write.
Arnavion 8y ago
>June 25, 2018, https://github.com/donnelldebnam/CodeU-Spring-2018-29/pull/59 https://github.com/donnelldebnam/CodeU-Spring-2018-29/pull/5... “Thanks!!” A strange fix to equally strange code. In the old code `tags` wasn't used in the first place, and in the new code it's put inside a scope ensuring it definitely can't be used. Why not just delete it? If it's being used just for side-effects in the `getAllHashtags()` call, why still bind it to a variable? Edit: Also, it isn't consistent with the original code in that it prefixes the member access with `this.` whereas the old code didn't do that. If I were the bot's author I wouldn't consider this commit a particularly good example of the bot's capability...
- devoply 8y ago
  The fixes as far as I can tell are pretty rudimentary, like if something is null and someone is trying to do something with it put an check for null around it. I would imagine if your used something like this consistently you code would become very much like a biological system -- a hack here a hack there... but hey it works.
chillacy 8y ago
Interesting idea, though the examples given were mainly fixing Null Pointer Exceptions, which can be somewhat solved with static analysis in languages without nullability.
CGamesPlay 8y ago
How does the actual "repairing" part work? Looks like it relies on a few external libraries and also includes custom code for dealing with NullPointerExceptions (which seem to be what was showcased here) - https://github.com/Spirals-Team/repairnator/tree/master/repairnator/repairnator-pipeline/src/main/java/fr/inria/spirals/repairnator/process/step/repair https://github.com/Spirals-Team/repairnator/tree/master/repa... - https://github.com/SpoonLabs/nopol https://github.com/SpoonLabs/nopol - https://github.com/SpoonLabs/astor https://github.com/SpoonLabs/astor

40 points by Dotnaught 8y ago

dang 8y ago

Url changed from https://www.theregister.co.uk/2018/10/17/luc_escape_bug_fixer/ https://www.theregister.co.uk/2018/10/17/luc_escape_bug_fixe..., which points to this.

jboggan 8y ago

I spent last year attempting something very similar in Python (www.duckly.com - unlaunched) and am very impressed. It's an extremely difficult problem and the CI data angle is inspired.

arayh 8y ago

I wonder if the Repairnator is in charge of writing its own commit comments (which I think is technically possible). The simple NullPointerException fixes don't look very impressive, although understandably still quite awesome for a bot. On the other hand, its commit on the eclipse/ditto repo was pretty interesting: https://github.com/eclipse/ditto/pull/151/commits/7ab96e373ecc55e1b942bfff4c9550c8ba5e584d https://github.com/eclipse/ditto/pull/151/commits/7ab96e373e... I'm quite surprised how it determined the usage of a function it did not write.

Arnavion 8y ago

>June 25, 2018, https://github.com/donnelldebnam/CodeU-Spring-2018-29/pull/59 https://github.com/donnelldebnam/CodeU-Spring-2018-29/pull/5... “Thanks!!” A strange fix to equally strange code. In the old code `tags` wasn't used in the first place, and in the new code it's put inside a scope ensuring it definitely can't be used. Why not just delete it? If it's being used just for side-effects in the `getAllHashtags()` call, why still bind it to a variable? Edit: Also, it isn't consistent with the original code in that it prefixes the member access with `this.` whereas the old code didn't do that. If I were the bot's author I wouldn't consider this commit a particularly good example of the bot's capability...

devoply 8y ago
The fixes as far as I can tell are pretty rudimentary, like if something is null and someone is trying to do something with it put an check for null around it. I would imagine if your used something like this consistently you code would become very much like a biological system -- a hack here a hack there... but hey it works.

chillacy 8y ago

Interesting idea, though the examples given were mainly fixing Null Pointer Exceptions, which can be somewhat solved with static analysis in languages without nullability.

CGamesPlay 8y ago

How does the actual "repairing" part work? Looks like it relies on a few external libraries and also includes custom code for dealing with NullPointerExceptions (which seem to be what was showcased here) - https://github.com/Spirals-Team/repairnator/tree/master/repairnator/repairnator-pipeline/src/main/java/fr/inria/spirals/repairnator/process/step/repair https://github.com/Spirals-Team/repairnator/tree/master/repa... - https://github.com/SpoonLabs/nopol https://github.com/SpoonLabs/nopol - https://github.com/SpoonLabs/astor https://github.com/SpoonLabs/astor