[llvm-dev] MemoryBuffer: Migrating to Expected/llvm::Error from ErrorOr/std::error_code

[David Blaikie via llvm-dev]

Hey folks - I'm/we're considering a somewhat laborious/disruptive refactoring for MemoryBuffer's APIs to improve error handling safety. Details here:  https://reviews.llvm.org/D109345

Duncan's pointed out that a reasonable migration strategy would be to:

1. Add MemoryBufferErrorAPI (wrapping APIs with errorOrToExpected) and MemoryBufferErrorCodeAPI (alias for MemoryBuffer) in the same commit.
2. Migrate in-tree callers to MemoryBufferErrorCodeAPI (via mass rename). (Could even move some to MemoryBufferErrorAPI?)
3. Update MemoryBuffer to use Error/Expected APIs, change MemoryBufferErrorAPI to an alias of it, and leave behind MemoryBufferErrorCodeAPI (wrapping APIs with expectedToErrorOr).
4. One or more commits:
   1. Migrate in-tree callers to MemoryBuffer.
   2. Delete MemoryBufferErrorAPI alias.
5. Delete MemoryBufferErrorCodeAPI wrappers.
   

(this isn't the only option (some variations discussed on the code review) - but something along these lines that separates the semantic changes from the renaming and makes it easier for folks with stable release branches to still cherry pick pieces of this without other pieces (eg: they can pick any amount of 1-4 without 5, specifically they could take 1-2, do a mass-rename on their branch of "MemoryBuffer:: -> MemoryBufferErrorCodeAPI::" and then be pretty stable/relatively easy to cherry pick things after that)

What do folks think of this? Any major objections (including "the churn doesn't seem worth the benefit" - happy to discuss that further), nuanced situations/tweaks/particular scenarios that we should make an effort to account for?

If there aren't huge objections - I'll probably start this in the next O(weeks).

[Lang Hames via llvm-dev]

I like this idea. MemoryBuffer is one of the biggest remaining users of ErrorOr (along with sys::FileSystem and VirtualFileSystem). Migrating MemoryBuffer to Expected would get us that much closer to eliminating ErrorOr entirely, which seems worth some churn.

-- Lang. 

[Jordan Rupprecht via llvm-dev]

Sounds like a good way to incrementally do a mass migration without disrupting downstream users so much, so LGTM.

[Fāng-ruì Sòng via llvm-dev]

Sounds good!
_______________________________________________
LLVM Developers mailing list
llvm...@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

[Mehdi AMINI via llvm-dev]

On Sun, Sep 19, 2021 at 2:44 PM David Blaikie via llvm-dev <llvm...@lists.llvm.org> wrote:

This looks like a reasonable plan, but what is the timeline for going from 1 to 5?
On top of the extra churn involved, I'd be wary of situations where the two APIs coexisted for too long, so hopefully O(weeks) and not O(months)?

 

If there aren't huge objections - I'll probably start this in the next O(weeks).

[David Blaikie via llvm-dev]

On Sat, Nov 20, 2021 at 4:37 PM Mehdi AMINI <joke...@gmail.com> wrote:

On Sun, Sep 19, 2021 at 2:44 PM David Blaikie via llvm-dev <llvm...@lists.llvm.org> wrote:

Hey folks - I'm/we're considering a somewhat laborious/disruptive refactoring for MemoryBuffer's APIs to improve error handling safety. Details here:  https://reviews.llvm.org/D109345

Duncan's pointed out that a reasonable migration strategy would be to:

1. Add MemoryBufferErrorAPI (wrapping APIs with errorOrToExpected) and MemoryBufferErrorCodeAPI (alias for MemoryBuffer) in the same commit.
2. Migrate in-tree callers to MemoryBufferErrorCodeAPI (via mass rename). (Could even move some to MemoryBufferErrorAPI?)
3. Update MemoryBuffer to use Error/Expected APIs, change MemoryBufferErrorAPI to an alias of it, and leave behind MemoryBufferErrorCodeAPI (wrapping APIs with expectedToErrorOr).
4. One or more commits:
   1. Migrate in-tree callers to MemoryBuffer.
   2. Delete MemoryBufferErrorAPI alias.
5. Delete MemoryBufferErrorCodeAPI wrappers.
   

(this isn't the only option (some variations discussed on the code review) - but something along these lines that separates the semantic changes from the renaming and makes it easier for folks with stable release branches to still cherry pick pieces of this without other pieces (eg: they can pick any amount of 1-4 without 5, specifically they could take 1-2, do a mass-rename on their branch of "MemoryBuffer:: -> MemoryBufferErrorCodeAPI::" and then be pretty stable/relatively easy to cherry pick things after that)

What do folks think of this? Any major objections (including "the churn doesn't seem worth the benefit" - happy to discuss that further), nuanced situations/tweaks/particular scenarios that we should make an effort to account for?

This looks like a reasonable plan, but what is the timeline for going from 1 to 5?
On top of the extra churn involved, I'd be wary of situations where the two APIs coexisted for too long, so hopefully O(weeks) and not O(months)?

Yeah, for sure a good thing to keep in mind. It is unfortunate to churn all the callers like this, but if we're going to do it, best to get back to the better/long term name sooner rather than later.