Re: [Hol-info] Share list of terms with later theories

[Jeremy Dawson]

A related question: some time back I was looking at how datatypes are 
constructed, and found stuff in theory ind_type, and theorems like 
list_TY_DEF (which one finds by doing find "ty_def")


But it seems that there are also theorems created which define the 
constructors of the list datatype in terms of ind_type$CONSTR, 
ind_type$BOTTOM, ind_type$FCONS, etc, but that these theorems are very 
well hidden.  I intended to investigate more to find how these theorems 
are hidden, and where if possible to find them, but never got back to it.


Can anyone help me on that question?  And is it relevant to this current 
discussion?


Cheers,

Jeremy


On 01/17/2018 10:28 AM, michael.norr...@data61.csiro.au wrote:

If people wanting to store these “uninteresting” theorems are happy to wrap and 
unwrap the OMITs, this would be one approach.

I had been thinking of adding a

   save_private_thm(name, privatedbname, thm)

entrypoint to Theory.sml.  You’d want multiple possible “private dbs”, so there 
would be an entry-point along the lines of

   get_private_db : string -> (string,thm) Binarymap.dict

Every time a theory was loaded, this private_db would change, so in many 
applications it might not be appropriate to store the result of

   get_private_db “cakeml/translator”

but to instead write

   val th = Binarymap.peek(get_private_db “cakeml/translator”, “theorem-name”)

Michael

On 16/1/18, 18:35, "Magnus Myreen"  wrote:

 Hi all,
 
 How about defining:
 
   OMIT x = x
 
 in markerScript.sml and making it print as ... and adjust HOL so that

 a theorem with a top-level OMIT does not show up in DB searches.
 
 Cheers,

 Magnus
 
 
 On 12 January 2018 at 00:00,   wrote:

 > I was thinking along these lines, yeah.  Such theorems could also be 
stopped from appearing in the Theory.sig file.
 >
 > Michael
 >
 > On 12/1/18, 07:31, "Konrad Slind"  wrote:
 >
 > Theorems that need to persist between sessions are most easily 
stored by name
 > in theories. Maybe some kind of PolyML database magic could also be
 > used, I don't
 > know. As far as DB searches, it wouldn't be hard to implement a
 > refined DB search
 > mechanism that first discarded all theorems that met some kind of
 > naming convention
 > (e.g., those starting with an underscore or something like that) and
 > then did the usual
 > search (which can be on name fragment or pattern).
 >
 > Konrad.
 >
 >
 > On Wed, Jan 10, 2018 at 7:09 PM, Magnus Myreen  
wrote:
 > > Ah, I didn't realise this existed. Thanks for the pointers!
 > >
 > > How does storing of theorems work in this setting? One can't 
construct
 > > a theorem from a string in a decode function.
 > >
 > > I guess the string could refer to a theorem name that's stored in 
the
 > > theory, but this is a bit inconvenient because some of the 
theorems in
 > > the translator's state are currently not stored in the theory 
(other
 > > than in the automatically produced theorem that is an encoding of 
the
 > > entire state). I guess an encode function could invent an unused 
name
 > > and store the theorem while it's producing the encoding. This can 
lead
 > > to some strange things turning up in DB searches (as is the case 
with
 > > the current approach).
 > >
 > > Cheers,
 > > Magnus
 > >
 > >
 > > On 11 January 2018 at 11:24,   
wrote:
 > >> That level of generality is already possible, and has always been 
a desideratum for the design.  (The grammar update information stored is about that 
complicated for example; consider the types that occur in a call to add_rule.)
 > >>
 > >> The painful part is that you have to write functions to encode 
and decode the types into and out of strings (because these strings are written into the 
theory files).  There are functions for doing basic SML types in src/parse/Coding, and 
the handling of terms is handled by writing functions that take functions for doing this 
as parameters.  See the bottom of src/parse/term_grammar for the encoding and decoding, 
and src/parse/GrammarDeltas for the way this is put together for the grammar example.
 > >>
 > >> You can see the fundamental building blocks at the 
LoadableThyData structure in src/postkernel/Theory.
 > >>
 > >> Certainly, providing a method for going through a generic 
s-expression type might be easiest for users to understand, so perhaps I can build that 
as well as term lists.
 > >>
 > >> Michael
 > >>
 > >> On 11/1/18, 11:08, "Magnus Myreen" 

Re: [Hol-info] Share list of terms with later theories

[Michael.Norrish]

If people wanting to store these “uninteresting” theorems are happy to wrap and 
unwrap the OMITs, this would be one approach.  

I had been thinking of adding a 

  save_private_thm(name, privatedbname, thm)

entrypoint to Theory.sml.  You’d want multiple possible “private dbs”, so there 
would be an entry-point along the lines of

  get_private_db : string -> (string,thm) Binarymap.dict

Every time a theory was loaded, this private_db would change, so in many 
applications it might not be appropriate to store the result of 

  get_private_db “cakeml/translator”

but to instead write

  val th = Binarymap.peek(get_private_db “cakeml/translator”, “theorem-name”)

Michael 

On 16/1/18, 18:35, "Magnus Myreen"  wrote:

Hi all,

How about defining:

  OMIT x = x

in markerScript.sml and making it print as ... and adjust HOL so that
a theorem with a top-level OMIT does not show up in DB searches.

Cheers,
Magnus


On 12 January 2018 at 00:00,   wrote:
> I was thinking along these lines, yeah.  Such theorems could also be 
stopped from appearing in the Theory.sig file.
>
> Michael
>
> On 12/1/18, 07:31, "Konrad Slind"  wrote:
>
> Theorems that need to persist between sessions are most easily stored 
by name
> in theories. Maybe some kind of PolyML database magic could also be
> used, I don't
> know. As far as DB searches, it wouldn't be hard to implement a
> refined DB search
> mechanism that first discarded all theorems that met some kind of
> naming convention
> (e.g., those starting with an underscore or something like that) and
> then did the usual
> search (which can be on name fragment or pattern).
>
> Konrad.
>
>
> On Wed, Jan 10, 2018 at 7:09 PM, Magnus Myreen  
wrote:
> > Ah, I didn't realise this existed. Thanks for the pointers!
> >
> > How does storing of theorems work in this setting? One can't 
construct
> > a theorem from a string in a decode function.
> >
> > I guess the string could refer to a theorem name that's stored in 
the
> > theory, but this is a bit inconvenient because some of the theorems 
in
> > the translator's state are currently not stored in the theory (other
> > than in the automatically produced theorem that is an encoding of 
the
> > entire state). I guess an encode function could invent an unused 
name
> > and store the theorem while it's producing the encoding. This can 
lead
> > to some strange things turning up in DB searches (as is the case 
with
> > the current approach).
> >
> > Cheers,
> > Magnus
> >
> >
> > On 11 January 2018 at 11:24,   
wrote:
> >> That level of generality is already possible, and has always been 
a desideratum for the design.  (The grammar update information stored is about 
that complicated for example; consider the types that occur in a call to 
add_rule.)
> >>
> >> The painful part is that you have to write functions to encode and 
decode the types into and out of strings (because these strings are written 
into the theory files).  There are functions for doing basic SML types in 
src/parse/Coding, and the handling of terms is handled by writing functions 
that take functions for doing this as parameters.  See the bottom of 
src/parse/term_grammar for the encoding and decoding, and 
src/parse/GrammarDeltas for the way this is put together for the grammar 
example.
> >>
> >> You can see the fundamental building blocks at the LoadableThyData 
structure in src/postkernel/Theory.
> >>
> >> Certainly, providing a method for going through a generic 
s-expression type might be easiest for users to understand, so perhaps I can 
build that as well as term lists.
> >>
> >> Michael
> >>
> >> On 11/1/18, 11:08, "Magnus Myreen"  wrote:
> >>
> >> Hi Michael,
> >>
> >> I see that you are considering to add a TermSetData feature. 
Could you
> >> please add something more general? I'd appreciate a feature 
that can
> >> store the CakeML translator's state in theories. Currently, 
the CakeML
> >> translator stores its state in a single theorem so that the 
other
> >> theories can load the state and continue from previous states.
> >>
> >> As far as I can tell, the ThmSetData machinery (and probably 
the
> >> TermSetData equivalent) won't help with the translator. The 
reason is
> >> that the translator's state is a collection of lists of list 
of tuples
> >>