SetupValue: Use proper TTG-style extension fields
#1919
Conversation
| module SAWScript.Crucible.Common.MethodSpec | ||
| ( AllocIndex(..) |
There was a problem hiding this comment.
Note to reviewers: this module still exports the same things that it did before this patch. I added an explicit export list here because this module now re-exports some things from SAWScript.Crucible.Common.Setup.Value, which requires using an export list to accomplish.
| module SAWScript.Crucible.JVM.MethodSpecIR | ||
| ( JIdent |
There was a problem hiding this comment.
Note to reviewers: this module still exports the same things that it did before this patch. I added an explicit export list here because this module now re-exports some things from SAWScript.Crucible.JVM.Setup.Value, which requires using an export list to accomplish.
| ppSetupValue :: Show (CastType ext) => SetupValue ext -> PP.Doc ann | ||
| -- Consider using 'resolveSetupValue' and printing the language-specific value | ||
| -- (e.g., an 'LLVMVal') with @PP.pretty@ instead. | ||
| ppSetupValue :: forall ext ann. IsExt ext => SetupValue ext -> PP.Doc ann |
There was a problem hiding this comment.
Adding this IsExt constraint to ppSetupValue is probably the largest change in this patch, aside from changing the TTG-related type families, as I have to perform case analysis on which extension is being used in order to pretty-print SetupValues properly. This is primarily what motivated moving code around so much in this patch, as I needed to be able to import all of the language backends from SAWScript.Crucible.Common.MethodSpec without creating import cycles.
| -- 'True' if this is an LLVM packed struct, 'False' otherwise. | ||
| type instance Setup.XSetupStruct (LLVM _) = Bool | ||
| type instance Setup.XSetupArray (LLVM _) = () | ||
| type instance Setup.XSetupElem (LLVM _) = () | ||
| type instance Setup.XSetupField (LLVM _) = () | ||
| type instance Setup.XSetupCast (LLVM _) = L.Type |
There was a problem hiding this comment.
These XSetupStruct and XSetupCast instances are the big payoffs in this patch. They no longer evaluate to (), but rather to Bool/L.Type. Moreover, this information is only used in the LLVM backend, which is why we are able to, e.g., delete so many uses of "packed" in the SetupStruct cases for the JVM and MIR backends.
This is purely a refactor. This will be useful in a subsequent commit to avoid import cycles.
…to *.Setup.Value modules This is purely a refactor. This will become useful in a subsequent commit in order to avoid import cycles.
This is purely a refactor. This will become useful in a subsequent commit in order to avoid import cycles.
This removes all of the `HasSetup* :: Bool` type families in favor of new `X* :: Type` type families. These largely serve the same purpose of distinguishing which constructors of `SetupValue` are permissible in each language backend, but they are permit bundling additional, language-specific information into a `SetupValue`. For instance, the `XSetupStruct` instance for `LLVM` evaluates to `Bool`, representing whether we are dealing with an LLVM packed struct or not, and the `XSetupCast` instance for `LLVM` evaluates to an LLVM `Type` to represent the type to cast to. One consequence of this design, aside from some `SetupValue` fields moving around, is that `ppSetupValue` now must perform case analysis on which language backend is currently in use in order to pretty-print any backend-specific data. This is accomplished with the new `SAWExt` data type and the corresponding `IsExt` class. There is a lot of code that shifts around in this patch, but everything is still functionally equivalent. In a subsequent patch, I will introduce additional data into MIR-specific extension fields. Fixes #1914.
I've included a pointer to #1929 as well to indicate our intention to support ghost state on other language backends.
RyanGlScott
added
the
PR: ready to merge
This removes all of the
HasSetup* :: Booltype families in favor of newX* :: Typetype families. These largely serve the same purpose of distinguishing which constructors ofSetupValueare permissible in each language backend, but they also permit bundling additional, language-specific information into aSetupValue. For instance, theXSetupStructinstance forLLVMevaluates toBool, representing whether we are dealing with an LLVM packed struct or not, and theXSetupCastinstance forLLVMevaluates to an LLVMTypeto represent the type to cast to.One consequence of this design, aside from some
SetupValuefields moving around, is thatppSetupValuenow must perform case analysis on which language backend is currently in use in order to pretty-print any backend-specific data. This is accomplished with the newSAWExtdata type and the correspondingIsExtclass.There is a lot of code that shifts around in this patch, but everything is still functionally equivalent. In subsequent patches, I will introduce additional data into MIR-specific extension fields.
Fixes #1914.