Refactor check_unique to Checked{Graph,Node}
#668
Conversation
kkysen
force-pushed
the
kkysen/pfb/unique-checking/checked-graph
branch
from
328afba to
449a7e3
Compare
kkysen
force-pushed
the
kkysen/pfb/unique-checking/check-unique-refactor
branch
from
34bc78f to
7510aec
Compare
…d running `.check_unique()`. It also preserves the Rust source code in debug info.
kkysen
force-pushed
the
kkysen/pfb/unique-checking/checked-graph
branch
from
449a7e3 to
ee941dd
Compare
|
|
||
| let a = g.mk_addr_of_local("let mut a = 0", false, 0_u32); | ||
| let b1 = g.mk_copy("let b = &mut a", false, a); | ||
| g.mk_store_addr("*b = 0", false, b1); |
There was a problem hiding this comment.
How about let _b2 = g.mk_store_addr(...), so the name b2 is visible here for matching with the graph above?
There was a problem hiding this comment.
Sure, I can add them back and just _-prepend them.
| fn mk_node( | ||
| &mut self, | ||
| stmt: &'static str, | ||
| unique: bool, |
There was a problem hiding this comment.
nit pick: i think i'd rather this be an enum than a bool
There was a problem hiding this comment.
optional change, just my two cents
There was a problem hiding this comment.
What's the advantage of an enum? If we're calling them Unique and NonUnique, I'm not sure there's an advantage over a bool. One is just non the other. Plus, then we should just change the unique field in NodeInfo::unique to the same enum.
There was a problem hiding this comment.
Plus, then we should just change the unique field in NodeInfo::unique to the same enum.
I agree that should be done. My preference is because I don't like relying on IDEs to tell me what the bool represents. A type Unique or NonUnique is very clear, despite it being semantically the same as a bool. In cases where functions take more than one bool (not the case here), it prevents mixing argument order by making things type safe.
There was a problem hiding this comment.
Okay, that makes sense. Plus we could change it to Unique and Aliases if we want, too. I do think we should save that for a later PR, though.
This refactors
check_uniquetoChecked{Graph,Node}.CheckedNodestores not just theNodeIdlike as passed to the previouscheck_unique, but also if it should be unique (thus we ensure we don't forget to check anyNodes (we did)), as well as the Rust source code statement for thatNodeas debug info.Then
CheckedGraphstores aGraphand aVec<CheckedNode>. When.mk_*methods are called on it, it adds theCheckedNodeto theVec, and then when.check_unique()is called at the end, it uses all theseCheckedNodes.The benefits of this are:
uniquevalue is right next to the call that creates thatNodeNodes have to be markeduniqueor not, so we don't forget any (we did)assert!ion fails and we print theNodeIdandGraphs, we also print the Rust source code as the debug info (previously this was left blank).