Changes between Version 3 and Version 4 of Holes
- Timestamp:
- 02/10/12 03:04:57 (15 months ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
Holes
v3 v4 1 1 = Holes = 2 2 3 One of the features of the Emacs mode for [http://wiki.portal.chalmers.se/agda/pmwiki.php Agda] is the ability to add goals . By inserting a {{{?}}} in an expression, the compiler will introduce a hole. After loading the file (which typechecks it), Agda gives an overview of the holes in the file and their types.3 One of the features of the Emacs mode for [http://wiki.portal.chalmers.se/agda/pmwiki.php Agda] is the ability to add goals, as a placeholder for code that is yet to be written. By inserting a {{{?}}} in an expression, the compiler will introduce a hole. After loading the file (which typechecks it), Agda gives an overview of the holes in the file and their types. 4 4 5 5 For example: … … 24 24 }}} 25 25 26 As can be seen here, holes are numbered, and the typechecker returns the name for each of these holes.26 As can be seen here, holes are numbered, and the typechecker returns the inferred type for each of these holes. 27 27 28 These goals can be useful as placeholders when writing code. They allow typechecking to continue although certain parts of code are missing and they make a goodTODO list.28 These goals can make it a lot easier to write code. They allow typechecking to continue although certain parts of code are missing and they work as a TODO list. 29 29 30 30 == In GHC == … … 36 36 As stated before, {{{undefined}}} typechecks just like a hole: it has type {{{a}}}, so it can be used anywhere. However, it is not very easy to use in this way: it is impossible to find out what type the compiler found for the hole, and it's impossible to get a list of all the holes used in your source file(s). 37 37 38 The sameexample:38 A similar example: 39 39 {{{ 40 40 test :: [Bool] … … 42 42 }}} 43 43 44 Will not help finding the types of the {{{undefined}}}s at all. One advantage is that the code will not refuse to run, unlessone of the {{{undefined}}}s is actually evaluated.44 Will not help finding the types of the {{{undefined}}}s at all. One advantage is that the code can successfully run, except if one of the {{{undefined}}}s is actually evaluated. 45 45 46 46 === Implicit Parameters === 47 The GHC extension [http://www.haskell.org/ghc/docs/latest/html/users_guide/other-type-extensions.html#implicit-parameters Implicit Parameters] comes closer to how we'd expect holes to work. It makes it possible to specify a term with a question mark, denoting a implicit variable.47 The GHC extension [http://www.haskell.org/ghc/docs/latest/html/users_guide/other-type-extensions.html#implicit-parameters Implicit Parameters] comes closer to how holes are expected to work. This extension makes it possible to specify a term with a question mark, denoting a implicit variable, but this could also be seen as a hole. 48 48 49 49 Same example: … … 79 79 }}} 80 80 81 This will show you the type, however, it does consider it an error and fails, so there may be other problems you don't get to see because of it. It also will refuse to load and compile the module, so it's impossible to run the parts of it that are finished.81 This will show you the type, however, it is an error and aborts compilation, so there may be other problems you don't get to see because of it. It also will refuse to load and compile the module, so it's impossible to run the parts of it that are finished. 82 82 83 83 The reason is that the hole becomes a part of the type signature, as a constraint. So to correctly use it here, the function would have to be written as: … … 88 88 }}} 89 89 90 This makes it very impractical to use them as holes, as all type signatures have to be updated to let the typechecker continue. Not only in the functions that use the implicit parameter itself, but they propagate upwards, just like class constraints: if another function were to call {{{test}}}, it would have the same implicit parameters (and therefore, all of these type signatures would have to be updated when a new hole is added). Another tricky problem with implicit parameters is that implicit parameters with the same name in different functions are not assumed to be the same parameter (i.e., required to be unifiable), ''except'' if some function has both implicit parameters in its constraints. Lastly, it's impossible to run code with unbound implicit parameters, even if the parameters are never actually used. 90 This makes it very impractical to use them as holes, as all type signatures have to be updated to let the typechecker continue. Not only in the functions that use the implicit parameter itself, but they propagate upwards, just like class constraints: if another function were to call {{{test}}}, it would have the same implicit parameters (and therefore, all of these type signatures would have to be updated when a new hole is added). 91 92 Another thing to keep in mind with implicit parameters is that implicit parameters with the same name in different functions are not assumed to be the same (i.e., required to be unifiable), ''except'' if some function has both implicit parameters in its constraints. Lastly, it's impossible to run code with unbound implicit parameters, even if the parameters are never actually used. 91 93 92 94 ---- 93 95 94 Now for possible ways we see holes working in GHC.96 Now for ways holes could be added to GHC(i). 95 97 96 98 === Agda-style === … … 112 114 113 115 === Named holes === 114 Implicit parameters have some good features too: they can be named, and so used in multiple locations. In the Agda-style, this would require let-binding a hole, which is a lot of effort for something that should be a temporary placeholder. So one idea is to allow giving holes a name, just like implicit parameters.116 Implicit parameters have some good features too: they can be named, and so used in multiple places. In the Agda-style, this would require let-binding a hole, which is a lot of effort for something that should be a temporary placeholder. So one idea is to allow giving holes a name, just like implicit parameters. 115 117 116 118 For example: … … 132 134 }}} 133 135 134 These could either be made shared within a module, or not (so not the confusing situation with implicit parameters, thatare only shared when required).136 These could either be made shared within all functions in module, or not shared between functions at all (so not the confusing situation with implicit parameters, which are only shared when required). 135 137 136 138 === Not holes, ranges === 137 Holes can be useful for finding the type of something that still needs to be written, but a more general way of looking at it is this: it is currently quite easy to typecheck a complete expression, for example with {{{:t}}}, in GHCi. However, finding the type of a part of an expression within a module is hard. In a large, complicated function it could be useful to ask the compiler for the types subexpressions. {{{:t}}} does not help here, as the function's parameters and where/let/lambda-bound terms are not in scope. It would be useful if it were possible to annotate code to ask the compiler to give you the type found for it.139 Holes can be useful for finding the type of something that still needs to be written, but a more general way of looking at it is this: it is currently quite easy to typecheck a complete expression, for example with {{{:t}}}, in GHCi. However, finding the type of a part of an expression within a module is hard. In a large, complicated function it could be useful to ask the compiler for the types of certain subexpressions. {{{:t}}} does not help here, as the function's parameters and where/let/lambda-bound terms are not in scope. It would be useful if it were possible to annotate code to ask the compiler to give you the type found for the annotated expression. 138 140 139 141 Simple example (let {{{{_ _}}}} denote a request for the type here):
