Re: Show instances for GHC internals

There are more interesting parts of your post, but I can respond to this: It shouldn't take that much time. Once you have ghc-stage2 built, you should be able to say `make 2` in the ./ghc subdirectory and get a new binary in a few seconds.

Using `make 1` in the ./compiler subdirectory works similarly for the stage1 compiler. But only once it's built the first time.

I replied this in other thread, I think it works but I'll make sure next time I do a `make clean`. Thanks. Another problem is this: Hiding fields of types is great for safety reasons, but not so great for debugging. In CoreDump I'm having this problems: - Sometimes GHC can't derive Show instance because record fields are hidden. But every field is actually exposed in a read-only way with some manually defined functions. This is super annoying. It'd be really awesome if we could export record fields as "read-only". (very half-baked idea) - Sometimes fields are hidden, and no accessors are provided. This is even worse because now there's really no way to derive Show, using `deriving` or manually.

(2) have unsafeInterleaveIO thunks which would induce IO action

Edward, do you remember any examples of such code? 2015-10-20 9:22 GMT-04:00 Ömer Sinan Ağacan :

...

One difficulty is that many of the core type data types, e.g. TyThing, are (1) a large mutually recursive graph, and (2) have unsafeInterleaveIO thunks which would induce IO action. So a naive Show instance would give infinite output and have lots of side effects. There are many data types which could usefully have Show added but also many which would be very difficult to.

Ah, yes, this is a very annoying problem. I discovered that when I first wrote CoreDump: https://github.com/osa1/CoreDump/issues/2

I don't have solution to this yet.

2015-10-19 21:59 GMT-04:00 Edward Z. Yang :

...

Excerpts from Ömer Sinan Ağacan's message of 2015-10-19 14:18:41 -0700:

...

I was wondering what would be the cost of adding Show instances. Would that mean significantly increased compile times? Or significantly bigger GHC binaries? If that's the case, could we enable Show instances with some arguments so that we can enable/disable it by modifying mk/build.mk?

One difficulty is that many of the core type data types, e.g. TyThing, are (1) a large mutually recursive graph, and (2) have unsafeInterleaveIO thunks which would induce IO action. So a naive Show instance would give infinite output and have lots of side effects. There are many data types which could usefully have Show added but also many which would be very difficult to.

Edward