Backend read #5004
Backend read #5004
Conversation
|
I'm happy to review this |
|
you closed the PR |
|
no merge? |
|
Ah, it didn't get merged. But in the github UI, it says "Pull request successfully merged"! |
|
And I can't re-open for some reason. I will re-PR from scratch. Sorry/thanks! |
|
https://github.com/hail-is/hail/commits/master not in commit history. Something happened though... |
|
wait, you force-pushed current master as this branch |
|
which reduced the diff to 0, and GitHub automatically closed? |
|
just back up in the reflog, you should be able to re-force-push and reopen? |
Yep, I put the wrong hash in a rebase --onto and when it was the same as master, Github closed it and marked it as merged. Surprising! Fixed! |
| @@ -10,21 +12,36 @@ class Backend(object): | |||
| def interpret(self, ir): | |||
| return | |||
|
|
|||
| @abc.abstractmethod | |||
There was a problem hiding this comment.
ah, nice. This will be the one entry point for types, and the rest will be handled in Python, right?
|
|
||
| assert isinstance(self._global_type, tstruct), self._global_type |
There was a problem hiding this comment.
no more assertions? I'm OK removing these, I suppose
There was a problem hiding this comment.
I put typecheck assertions on the ttable/tmatrix constructors. I thought that was good enough.
Builds on: #4995
First of several (many) changes to rip out Java dependence from the mainline Python code. Move _to_java_ir to the SparkBackend. Eventually, only the Spark backend should call Java. Matrix/Table now gets it types from BaseIR.typ (which will eventually implement the (virtual) typing rules in Python). Matrix/Table explicitly call into the Spark backend to create _jmt/_jt, but those can be removed once the methods are all using IR.
@tpoterba giving this to you since you got the last one. Let me know if you would prefer me to spin the wheel.