Add a lazyt class for delaying computations #5090
Conversation
romainbrenguier
requested review from
kroening,
peterschrammel,
pkesseli,
smowton and
tautschnig
as code owners
romainbrenguier
force-pushed
the
feature/lazy
branch
from
1fed884 to
e5bf82d
Compare
There was a problem hiding this comment.
✔️
Passed Diffblue compatibility checks (cbmc commit: e5bf82d).
Build URL: https://travis-ci.com/diffblue/test-gen/builds/126077346
Codecov Report
@@ Coverage Diff @@
## develop #5090 +/- ##
===========================================
+ Coverage 69.66% 69.66% +<.01%
===========================================
Files 1319 1321 +2
Lines 109250 109269 +19
===========================================
+ Hits 76106 76125 +19
Misses 33144 33144
Continue to review full report at Codecov.
|
src/util/lazy.h
Outdated
|
|
||
| /// Force the computation of the value. If it was already computed (even if | ||
| /// \p info was different), return the same result. | ||
| valuet force(infot info) |
There was a problem hiding this comment.
I find "force" an unintuitive name. How about "evaluate", "compute" or "force_evaluation"?
There was a problem hiding this comment.
force is used by .net, scheme and ocaml https://en.wikipedia.org/wiki/Lazy_evaluation so I think it's relatively standard, but I don't know if someone encountered similar concepts in other libraries or languages?
There was a problem hiding this comment.
If it's a standard term then that's fine.
| static lazyt from_fun(std::function<valuet(infot)> fun) | ||
| { | ||
| return lazyt{std::move(fun)}; | ||
| } |
There was a problem hiding this comment.
Why not a public constructor?
There was a problem hiding this comment.
I think having a static function with a name is clearer in that case, when I read lazyt::from_fun(f) I know we use a function f to make a lazyt object, but lazyt(f) is more obscure.
This generalizes a pattern for which we have several potential usage.
This tests that the force function returns the expected result and the evalution function is not called more than necessary.
romainbrenguier
force-pushed
the
feature/lazy
branch
from
e5bf82d to
8dbbca1
Compare
|
@owen-mc-diffblue @danpoe following the suggestion of @LAJW I've removed the |
There was a problem hiding this comment.
✔️
Passed Diffblue compatibility checks (cbmc commit: 8dbbca1).
Build URL: https://travis-ci.com/diffblue/test-gen/builds/126233098
| return s.length(); | ||
| }; | ||
| auto lazy_length = | ||
| lazyt<int>::from_fun([&]() { return length_with_counter("foo"); }); |
There was a problem hiding this comment.
😊 https://docs.microsoft.com/en-us/dotnet/fsharp/language-reference/lazy-expressions
And now you could even infer return type by doing:
template<T> auto lazy(T func) -> lazyt<decltype(func())> { ... }
Also, () in zero-argument lambdas is optional.
There was a problem hiding this comment.
I'm adding it in a follow-up PR: #5101
This class is meant to be used for delaying the computation of a value to the first time we need it, but avoid recomputing it if it is needed several times.
In particular I would have found such a class useful when writing this pull request: #5078