When I explain to people what strict/lazy/eager mean, I often say something like :
- Adjectives eager and lazy apply *only* to a global evaluation method: * eager* is C evaluation style and *lazy* is that of Haskell. - Adjective strict can be applied *both* to a global evaluation method and a specific function: if applied to an eval method then it's a synonym of "strict", and if applied to a function f it means 'f ⊥ = ⊥' (which I detail a little more), which is true for strict State monad for istance (>>= would not allow its left argument to return ⊥). Thus explaining why datatypes such as State or Bytestring exist in *strict * and* lazy *flavours. 2011/12/28 Albert Y. C. Lai <tre...@vex.net> > There are two flavours of MonadState, Control.Monad.State.Lazy and > Control.Monad.State.Strict. There are two flavours of ByteString, > Data.ByteString.Lazy and Data.Bytestring (whose doc says "strict"). There > are two flavours of I/O libraries, lazy and strict. There are advices of > the form: "the program uses too much memory because it is too lazy; try > making this part more strict". Eventually, someone will ask what are "lazy" > and "strict". Perhaps you answer this (but there are other answers, we'll > see): > > "lazy refers to such-and-such evaluation order. strict refers to f ⊥ = ⊥, > but it doesn't specify evaluation order." > > That doesn't answer the question. That begs the question: Why do libraries > seem to make them a dichotomy, when they don't even talk about the same > level? And the make-it-more-strict advice now becomes: "the program uses > too much memory because of the default, known evaluation order; try making > this part use an unknown evaluation order", and this unknown is supposed to > use less memory because...? > > I answer memory questions like this: the program uses too much memory > because it is too lazy---or nevermind "lazy", here is the current > evaluation order of the specific compiler, this is why it uses much memory; > now change this part to the other order, it uses less memory. I wouldn't > bring in the denotational level; there is no need. > > (Sure, I use seq to change evaluation order, which may be overriden by > optimizations in rare cases. So change my answer to: now add seq here, > which normally uses the other order, but optimizations may override it in > rare cases, so don't forget to test. Or use pseq.) > > I said "people, make up your mind". I do not mean a whole group of people > for the rest of their lives make up the same mind and choose the same one > semantics. I mean this: Each individual, in each context, for each problem, > just how many levels of semantics do you need to solve it? (Sure sure, I > know contexts that need 4. What about daily programming problems: time, > memory, I/O order?) > > MigMit questioned me on the importance of using the words properly. > Actually, I am fine with using the words improperly, too: "the program uses > too much memory because it is too lazy, lazy refers to such-and-such > evaluation order; try making this part more strict, strict refers to > so-and-so evaluation order". > > > > ______________________________**_________________ > Haskell-Cafe mailing list > Haskell-Cafe@haskell.org > http://www.haskell.org/**mailman/listinfo/haskell-cafe<http://www.haskell.org/mailman/listinfo/haskell-cafe> >
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
