gobo-eiffel-develop — Development discussions

Re: [gobo-eiffel-develop] Multi-processing whilst compiling with gec

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
> I notice that whereas ise will launch multiple jobs to compile
> generated C code, gec appears to only launch one - resulting in long
> compile times.
> 
> Is this inherent in the process, or is it due to my command line:

No work has been done to let gec compile several C files
concurrently. Currently it is a mere script compiling the
C files one after the other.

No work has been done either on trying to make the C code
generation incremental, and hence avoiding having to recompile
C code that has not changed between two successive compilations.

And no effort has been made to try to make the generated C
code as optimized as possible.

There are plenty of areas where gec can be improved. Which
means that there is room to make gec even better in the
near future. Currently my focus is more on the Eiffel part,
trying to make it more compliant with ISE Eiffel and more
generally with ECMA/ISO Eiffel.

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

[gobo-eiffel-develop] Multi-processing whilst compiling with gec

[Colin P. A. <co...@co...>]

I notice that whereas ise will launch multiple jobs to compile
generated C code, gec appears to only launch one - resulting in long
compile times.

Is this inherent in the process, or is it due to my command line:

gec --finalize --cc=yes --catcall=error --gc=boehm compile_ge.ace

-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Colin" == Colin Adams <col...@go...> writes:

    Colin> OK. I understand your point now.  I did think at the
    Colin> weekend about making it a generic parameter (constrained to
    Colin> a descendant of NUMERIC), but then I got side-tracked by
    Colin> the possibility of eliminating MA_DECIMAL altogether.

    Colin> I will investigate this at the weekend.

I've reviewed DT_XSLT_NUMBERER_EN to see whether this is possible.

The main problem seems to be the use of
{MA_DECIMAL}.is_integer/to_integer as NUMERIC doesn't have these.

The use of {MA_DECIMAL}.to_scientific_string is also an issue - but it
is probably non-conformant as it is.

I can eliminate MA_DECIMAL quite easily by using a new deferred class
which will have two descendants, one based on INTEGER_64 or
INTEGER_32, and one based on MA_DECIMAL.

But you are already unhappy about the use of auxiliary classes.

And I definitely can't eliminate the unicode library usage.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Colin A. <col...@go...>]

On 07/01/2008, Eric Bezault <er...@go...> wrote:
>
> Colin Adams wrote:
>
> > Unfortunately, yes, it is (I did have a look into this).
> > What is more, it will get worse in the future - so far there is only an
> > English language version. There will be more.
>
> What impact would it have on the integers held in a date?
> Would they use other unicode characters for the digits?


That is already possible even for the English language version.
The main significance of language is if you choose  word output.
E.g. Two Thousand and Seven (rather than 2007).


>     It sounds to me as if because when written to a file an integer
> >     and a decimal look the same we should first convert INTEGERs to
> >     MA_DECIMALs when passing them to class FILE.
> >
> >
> > Now you've lost me.
>
> In other words, for anything I want to do in an INTEGER, why
> would I need to convert it to a MA_DECIMAL first? If I dpn't
> do this conversion for anything I do with an INTEGER, why should
> it be different in this particular case?



OK. I understand your point now.
I did think at the weekend about making it a generic parameter (constrained
to a descendant of NUMERIC), but then I got side-tracked by the possibility
of eliminating MA_DECIMAL altogether.

I will investigate this at the weekend.

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Adams wrote:
> On 07/01/2008, *Eric Bezault* <er...@go... 
> <mailto:er...@go...>> wrote:
> 
>     Colin Adams wrote:
> 
>      > But the code is shared by xsl:number (as the formatting rules are
>      > defined to be the same), which might format any number.
> 
>     And is it that much code duplication to implement the formatting
>     rules both for an INTEGER and for a MA_DECIMAL as input?
> 
> 
> Unfortunately, yes, it is (I did have a look into this).
> What is more, it will get worse in the future - so far there is only an 
> English language version. There will be more.

What impact would it have on the integers held in a date?
Would they use other unicode characters for the digits?

>     It sounds to me as if because when written to a file an integer
>     and a decimal look the same we should first convert INTEGERs to
>     MA_DECIMALs when passing them to class FILE.
> 
> 
> Now you've lost me.

In other words, for anything I want to do in an INTEGER, why
would I need to convert it to a MA_DECIMAL first? If I dpn't
do this conversion for anything I do with an INTEGER, why should
it be different in this particular case?

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Date/time formatting

[Colin A. <col...@go...>]

On 07/01/2008, Eric Bezault <er...@go...> wrote:
>
> Colin Adams wrote:
>
> > But the code is shared by xsl:number (as the formatting rules are
> > defined to be the same), which might format any number.
>
> And is it that much code duplication to implement the formatting
> rules both for an INTEGER and for a MA_DECIMAL as input?


Unfortunately, yes, it is (I did have a look into this).
What is more, it will get worse in the future - so far there is only an
English language version. There will be more.


It sounds to me as if because when written to a file an integer
> and a decimal look the same we should first convert INTEGERs to
> MA_DECIMALs when passing them to class FILE.


Now you've lost me.

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Adams wrote:
> 
> 
> On 07/01/2008, *Eric Bezault* <er...@go... 
> <mailto:er...@go...>> wrote:
> 
> 
>     BTW, how come we need such big numbers to format dates?
> 
> We don't.
> But the code is shared by xsl:number (as the formatting rules are 
> defined to be the same), which might format any number.

And is it that much code duplication to implement the formatting
rules both for an INTEGER and for a MA_DECIMAL as input?

It sounds to me as if because when written to a file an integer
and a decimal look the same we should first convert INTEGERs to
MA_DECIMALs when passing them to class FILE. Code duplication is
one thing, but wasting execution time to convert entities in one
way or another just for the sake of avoiding code duplication
(and introducing library dependencies during that process) in
another thing that is not necessarily better.

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Date/time formatting

[Colin A. <col...@go...>]

On 07/01/2008, Eric Bezault <er...@go...> wrote:
>
>
> BTW, how come we need such big numbers to format dates?
>
> We don't.
But the code is shared by xsl:number (as the formatting rules are defined to
be the same), which might format any number.

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
> I got my requested clarification from the working group - the limit
> only applies if the range of xs:integer is limited by the
> implementation.
> And it was to avoid doing this that we introduced MA_DECIMAL into Gobo
> in the first place.

BTW, how come we need such big numbers to format dates?

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
> I got my requested clarification from the working group - the limit
> only applies if the range of xs:integer is limited by the
> implementation.
> And it was to avoid doing this that we introduced MA_DECIMAL into Gobo
> in the first place.

I don't think that MA_DECIMAL has been introduced into Gobo to
accommodate xs:integer ;-)

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Colin" == Colin Paul Adams <co...@co...> writes:

    Colin> But after looking at the appendix to the XSLT
    Colin> recommendation, I see that an implementation is allowed to
    Colin> specify a maximum number that can be formatted. And as I
    Colin> can't see any practical use cases for formatting larger
    Colin> numbers than can be represented by INTEGER (the intended
    Colin> use case for xsl:number is for formatting section
    Colin> numbering, and the like), I shall have a go at eliminating
    Colin> the use of MA_DECIMAL.

I got my requested clarification from the working group - the limit
only applies if the range of xs:integer is limited by the
implementation.
And it was to avoid doing this that we introduced MA_DECIMAL into Gobo
in the first place.

So it will have to stay.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Eric" == Eric Bezault <er...@go...> writes:

    >> Well, I won't be able to remove any dependency on structure

    Eric> Do you know where this dependency on structure comes from?

 DT_XSLT_FORMAT_DATE_TIME:

Lots of use of DS_CELL for passing arguments. These could be changed
to KL_CELL.
Two uses of ST_SPLITTER/DS_LIST for parsing values (that seems to be
the string library).
Since the splitting is only on a single character, I guess it wouldn't
be a big job after all to remove this, BUT:

DT_XSLT_NUMBER_ROUTINES:
 This has string/unicode dependencies only - these are vital, and
 can not be removed.
And including this library is a much larger overhead than anything
 else we have been talking about.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
>>>>>> "Eric" == Eric Bezault <er...@go...> writes:
> 
>     >> Well this isn't new. If you look at the system.xace in the
>     >> test/time directory, you will see quite a few dependencies.
>     >> Mind you, some of them may just be for the test framework.
> 
>     Eric> Yes, there are for the test framework. I had to add them in
>     Eric> example/time/clock.
> 
> Well, I won't be able to remove any dependency on structure 

Do you know where this dependency on structure comes from?

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Eric" == Eric Bezault <er...@go...> writes:

    >> Well this isn't new. If you look at the system.xace in the
    >> test/time directory, you will see quite a few dependencies.
    >> Mind you, some of them may just be for the test framework.

    Eric> Yes, there are for the test framework. I had to add them in
    Eric> example/time/clock.

Well, I won't be able to remove any dependency on structure (so I
might as well forget about removing the dependency on math, although I
will still do it if I can, as it slows down execution a lot).
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
>>>>>> "Eric" == Eric Bezault <er...@go...> writes:
> 
>     Eric> Colin Paul Adams wrote:
>     >> I shall have a go at eliminating the use of MA_DECIMAL.
> 
>     Eric> Great. Also note that my experience, running the Gobo
>     Eric> bootstrap and test suite with ISE, showed that in addition
>     Eric> to regexp and math, the date/time library now also depends
>     Eric> on library/string and library/structure.
> 
> Well this isn't new. If you look at the system.xace in the test/time
> directory, you will see quite a few dependencies. 
> Mind you, some of them may just be for the test framework.

Yes, there are for the test framework. I had to add them in
example/time/clock.

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Unicode regular expressions

[Colin P. A. <co...@co...>]

>>>>> "Eric" == Eric Bezault <er...@go...> writes:

    Eric> Colin Paul Adams wrote:
    >> Here is advanced notice that I have started preliminary work
    >> (since Christmas, actually) on a regular expression library for
    >> use with Unicode text. It will support muliple dialects for
    >> regular expression languages (the initial implementation will
    >> support the XPath and XML Schema regular expression
    >> languages. My need is for the former, but it is a superset of
    >> the latter, so I might as well do both at once).
    >> 
    >> It will be at least a month before I have anything worth
    >> looking at.

    Eric> OK. Does that mean that there will be a new dependency on
    Eric> library/string to have access to its unicode cluster?

Well, that will depend where we decide to put it.

Since I am messing about with the design a lot, I wanted to check it
in to experimental (just so I can look back at my experiments in case
I delete something and decide I want it after all). But this doesn't
exist in the svn version. So I'm checking in to the gestalt
sourceforge site (as that is a set of programs rather than a library,
it isn't a problem for me).

Then when I have something working, I shall ask for a preliminary
assesment, comments on the approach (I shall be using 0-127 - as
indices into a state transition for ASCII characters, as 128+ to
represent character classes. Hopefully that way we can maintain
performance for ASCII, whilst still accomodaing full Unicode
compatibility, and reasonable size tables) and design (does anyone
know of any papers in this area?).

After that, I'll do any necessary re-working, and we can decide on a
library structure.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Eric" == Eric Bezault <er...@go...> writes:

    Eric> Colin Paul Adams wrote:
    >> I shall have a go at eliminating the use of MA_DECIMAL.

    Eric> Great. Also note that my experience, running the Gobo
    Eric> bootstrap and test suite with ISE, showed that in addition
    Eric> to regexp and math, the date/time library now also depends
    Eric> on library/string and library/structure.

Well this isn't new. If you look at the system.xace in the test/time
directory, you will see quite a few dependencies. 
Mind you, some of them may just be for the test framework.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Unicode regular expressions

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
> Here is advanced notice that I have started preliminary work (since
> Christmas, actually) on a regular expression library for use with
> Unicode text. It will support muliple dialects for regular expression
> languages (the initial implementation will support the XPath and XML
> Schema regular expression languages. My need is for the former, but it
> is a superset of the latter, so I might as well do both at once).
> 
> It will be at least a month before I have anything worth looking at.

OK. Does that mean that there will be a new dependency on
library/string to have access to its unicode cluster?

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Colin" == Colin Paul Adams <co...@co...> writes:

    Colin> But after looking at the appendix to the XSLT
    Colin> recommendation, I see that an implementation is allowed to
    Colin> specify a maximum number that can be formatted.

Unfortunately this appendix is non-normative. There may or may not be
a contradiction with the body of the recommendation. I am seeking a
ruling from the working group.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
>  I shall have a go at eliminating the use of MA_DECIMAL.

Great. Also note that my experience, running the Gobo bootstrap
and test suite with ISE, showed that in addition to regexp and
math, the date/time library now also depends on library/string
and library/structure.

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Eric" == Eric Bezault <er...@go...> writes:

    Eric> Colin Paul Adams wrote:
    >>>>>>> "Colin" == Colin Paul Adams <co...@co...>
    >>>>>>> writes:
    >> 
    Colin> I can eliminate the regular expression without too much
    Colin> trouble.
    >> 
    >> I've done that now.

    Eric> OK, thanks.

    Eric> Note that I had already started the process necessary to
    Eric> check-in the snapshot of Gobo at AXAR. Do you want me to
    Eric> start again from scratch? It's not a big deal for me, it's
    Eric> just that it would delay the process by one day or so. Just
    Eric> let me know.

No it doesn't make any difference. The next check-in will just remove
the dependency silently, and won't make any difference.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Eric B. <er...@go...>]

Colin Paul Adams wrote:
>>>>>> "Colin" == Colin Paul Adams <co...@co...> writes:
> 
>     Colin> I can eliminate the regular expression without too much
>     Colin> trouble.
> 
> I've done that now.

OK, thanks.

Note that I had already started the process necessary to check-in
the snapshot of Gobo at AXAR. Do you want me to start again from
scratch? It's not a big deal for me, it's just that it would delay
the process by one day or so. Just let me know.

-- 
Eric Bezault
mailto:er...@go...
http://www.gobosoft.com

[gobo-eiffel-develop] Unicode regular expressions

[Colin P. A. <co...@co...>]

Here is advanced notice that I have started preliminary work (since
Christmas, actually) on a regular expression library for use with
Unicode text. It will support muliple dialects for regular expression
languages (the initial implementation will support the XPath and XML
Schema regular expression languages. My need is for the former, but it
is a superset of the latter, so I might as well do both at once).

It will be at least a month before I have anything worth looking at.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Colin" == Colin Paul Adams <co...@co...> writes:

    Colin> Eliminating the use
    Colin> of MA_DECIMAL will be much harder - it can be done, but not
    Colin> without duplicating a lot of code.

The need was because the code is also used by the implementation of
xsl:number (formatting dates and times are already limited to INTEGER
by virtue of DT_DATE using INTEGER for the year).

But after looking at the appendix to the XSLT recommendation, I see
that an implementation is allowed to specify a maximum number that can
be formatted. And as I can't see any practical use cases for
formatting larger numbers than can be represented by INTEGER (the
intended use case for xsl:number is for formatting section numbering,
and the like), I shall have a go at eliminating the use of MA_DECIMAL.
-- 
Colin Adams
Preston Lancashire

Re: [gobo-eiffel-develop] Date/time formatting

[Colin P. A. <co...@co...>]

>>>>> "Colin" == Colin Paul Adams <co...@co...> writes:

    Colin> I can eliminate the regular expression without too much
    Colin> trouble.

I've done that now.
-- 
Colin Adams
Preston Lancashire