[Adding ghc-devs@haskell.org to cc] On 21 May 2015 at 02:12, Sergei Meshveliani wrote:

Now, I delete `OverlappingInstances' from docon.cabal

and also from the $doconCpOpt options to call ghc on demotest/Main.hs.

And now the test runs correct in ghc-7.10.1 !

Only it is 1.5 times slower than in ghc-7.8.2. So: a) The test intends overlapping instances, b) instance overlaps are not declared for ghc-7.10.1, c) this leads to a correct running, but 1.5 times slower.

I do not know of whether this slow down is due to the removed pragma or due to other features of ghc-7.10.1.

And there remains my question about how GHC could help to recall the overlapping instance pairs.

Regards,

------ Sergei

On Wed, 2015-05-20 at 23:52 +0400, Sergei Meshveliani wrote:

...

Dear GHC developers,

Please, test ghc-7.10.1 on making docon-2.12

http://www.botik.ru/pub/local/Mechveliani/docon/

and running its demotest/Main (see install.txt).

docon-2.12 has been tested under ghc-7.8.2,

and it has extensions: ... OverlappingInstances in docon.cabal.

Making with ghc-7.10.1 issues a warning and advises to set the related pragma individually to each corresponding instance.

(1) First, I ignored this warning, and surprisingly, the library has been made.

In the build/ subdirectory there are .hi and .o files, and there have newly appeared the "dyn" files:

Matr0_.dyn_hi Matr0_.dyn_o Matr0_.hi Matr0_.o ...

(I do not know what does it mean "dyn").

Then, > make install reports

runghc Setup.hs install --user Installing library in

/home/mechvel/docon/2.12/docon/source/inst/lib/x86_64-linux-ghc-7.10.1/docon_99cUeE74HI58Sb9XilYAZ2

...

Registering docon-2.12.1...

Instead of the .a file, ls shows there

docon_99cUeE74HI58Sb9XilYAZ2 docon_JzkSttB1MsX3R5AQ1eIgvC

Are these names intended?

Then I command

...

cd demotest ghc $doconCpOpt -O -rtsopts --make Main ./Main

It is built and runs. But breaks in the middle with a certain reasonable DoCon error message.

The DoCon design is so that choosing a different instance among the overlapping ones may change the algorithm, and the computation cost, but still must produce the same result. It must -- unless DoCon has a bug in some of these instances.

(2) All right, I need to set {-# OVERLAPPABLE #-} individually to each instance which may overlap with some other instance in DoCon.

There is declared a great number of instances. And it occurs difficult for me to recall now: which instance does overlap with something and which does not (at least they all worked correct in ghc-7.8.2).

Well, I can set {-# OVERLAPPABLE #-} to _each_ instance. But such a program does not look nice.

And I would like to be more definite and to set {-# OVERLAPPING #-} to all appropriate places. Suppose that I forget that some instance overlaps with something and skipped this pragma. The above test shows that this can lead to a wrong program to run (is this feature intended?).

And what will be the consequence if {-# OVERLAPPING #-} is set to an instance which does not overlap with anything?

Can the compiler help to list the overlaps?

Please, advise,

------ Sergei

_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/glasgow-haskell-users

_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/glasgow-haskell-users

-- Regards Sumit Sahrawat