Objectada link time

Objectada link time

[Nicolas Brunot]

Consulting FAQ on Aonix web site, I found the following :

>>5.My compiler seems slow. How can I speed it up?
>>...
>>Sometimes users report a slow build time. Usually, much of the build
time
>>occurs during the link phase, not during the compilation phase, a
reflection of
>>the architecture of the Microsoft linker.

Comparing with Rational Apex which also include the Microsoft linker,
there is a enormous difference.
for an application of about 15 Mo, linked from about 1200 Ada units :
Apex link is about 1 mn
Objectada link reaches 10 mn ...

So there should be not only a microsoft linker problem here.

Has anybody found a solution to this kind of problem ?
Thanks for any information

Re: Objectada link time

[Nicolas Brunot]

Testing link time in /VERBOSE mode, it appears that after the message
"Start Pass1"
it takes several minutes to Objectada before the following verbose
messages appear, starting by
"Searching Libraries"
etc ...

On others compilers, this is quite instantaneous.
No idea of what could happen between
"Start Pass1"
and
"Searching Libraries" ?

Re: Objectada link time

[John Magness]

Nicolas Brunot wrote:

> Consulting FAQ on Aonix web site, I found the following :
>
> >>5.My compiler seems slow. How can I speed it up?
> >>...
> >>Sometimes users report a slow build time. Usually, much of the build
> time
> >>occurs during the link phase, not during the compilation phase, a
> reflection of
> >>the architecture of the Microsoft linker.
>
> Comparing with Rational Apex which also include the Microsoft linker,
> there is a enormous difference.
> for an application of about 15 Mo, linked from about 1200 Ada units :

Mo? is this megabytes of object code? What are the units?

>
> Apex link is about 1 mn
> Objectada link reaches 10 mn ...
>
> So there should be not only a microsoft linker problem here.
>

 What platforms produced these numbers?  The last line might lead a person
to think
that the same platform MS/Intel was used. A person familiar with the
internals of Apex might be able to throw some  light on this.

>
> Has anybody found a solution to this kind of problem ?
> Thanks for any information

This is a curious situation....
John

Re: Objectada link time

[Keith Thompson]

John Magness <jmagness@swbell.net> writes:
> Nicolas Brunot wrote:
[...]
> > Comparing with Rational Apex which also include the Microsoft linker,
> > there is a enormous difference.
> > for an application of about 15 Mo, linked from about 1200 Ada units :
> 
> Mo? is this megabytes of object code? What are the units?

Probably mega-octets, what most of us sloppily call megabytes.
(An octet is unambiguously 8 bits; a byte is *usually* 8 bits,
but it can vary.)

-- 
Keith Thompson (The_Other_Keith) kst@cts.com  <http://www.ghoti.net/~kst>
San Diego Supercomputer Center           <*>  <http://www.sdsc.edu/~kst>
Welcome to the last year of the 20th century.

Re: Objectada link time

[Nicolas Brunot]

John Magness wrote

> >> Comparing with Rational Apex which also include the Microsoft linker,
> >> there is a enormous difference.
> >> for an application of about 15 Mo, linked from about 1200 Ada units :
>
> >Mo? is this megabytes of object code? What are the units?

there are about 650 Ada packages giving a .rsp file of about 1300 lines (about
1300 .obj files for Ada specs and bodies)
15 Mo (mega-octets) is the size of the executable generated

> >>
> >> Apex link is about 1 mn
> >> Objectada link reaches 10 mn ...
> >>
> >> So there should be not only a microsoft linker problem here.
> >>
>
> > What platforms produced these numbers?  The last line might lead a person
> >to think
> >that the same platform MS/Intel was used. A person familiar with the
> >internals of Apex might be able to throw some  light on this.

Microsoft Windows NT4, biprocessor Pentium II 400 with 256 Mo for RAM

>

Re: Objectada link time

[Robert A Duff]

Keith Thompson <kst@cts.com> writes:

> Probably mega-octets, what most of us sloppily call megabytes.
> (An octet is unambiguously 8 bits; a byte is *usually* 8 bits,
> but it can vary.)

You're showing your age.  ;-) ;-)

- Bob

Re: Objectada link time

[John Magness]

Nicolas Brunot wrote:

> Apex link is about 1 mn
> Objectada link reaches 10 mn ...
>
> So there should be not only a microsoft linker problem here.
>
> Has anybody found a solution to this kind of problem ?
> Thanks for any information

Just a guess, but are the compiler options selected logically equivalent?
--- I believe I've read that ObjectAda does no optimization....
inclusion/exclusion of debug data?

Octets?  How about 36 bit words (Univac) and 24 bit words (Harris, I seem
to recall)?
John

Re: Objectada link time

[Nicolas Brunot]

John Magness a �crit :

> >> Apex link is about 1 mn
> >> Objectada link reaches 10 mn ...
> >>
> >> So there should be not only a microsoft linker problem here.
> >>
> >> Has anybody found a solution to this kind of problem ?
> >> Thanks for any information
>
> >Just a guess, but are the compiler options selected logically equivalent?
> >--- I believe I've read that ObjectAda does no optimization....
> >inclusion/exclusion of debug data?
>

Hard to know if they have the same effect, and actually total size of object
files is bigger with objectada than with apex.
But I don't think objects file total size or compilers options is the
problem.
No matter which options you use, Apex link is always by far much faster than
Objectada
It wouldn't explain such a difference (link about 10 times faster than
Objectada)

in /VERBOSE mode, almost all the time difference occurs between the 1rst and
2nd line of linker output
"Start Pass1"
and
"Searching Libraries"
and the linker output is about 13000 lines !!

Re: Objectada link time

[Gautier]

Nicolas Brunot:

> It wouldn't explain such a difference (link about 10 times faster than
> Objectada)
>
> in /VERBOSE mode, almost all the time difference occurs between the 1rst and
> 2nd line of linker output
> "Start Pass1"
> and
> "Searching Libraries"
> and the linker output is about 13000 lines !!

Ouch! They could hire a programmer from ex-Borland to improve this - or take
a look at the disassembled Turbo Pascal 6 sources and especially its fast
smart linker...

G.