I think object persistence is the last missing feature that keep me from jumping to D.  Walter, what's your roadmap for this?


BTW, follow the D link from (http://www.digitalmars.com/d/index.html)

http://www.functionalfuture.com/d/

why D is so bad on the follwing 4 tests in the middle? especially the hashes?

MSVC7  GCC    D      C#
fibo (42)       6.597  6.762  5.910  8.051
-------------------------------------------
ackermann (12)  3.005  2.929  6.064  9.017 hash (1000000)  1.954  2.653 14.929  3.013 hash2 (2000)    4.440  4.059 32.162 21.041 matrix (100000) 7.039  7.051 16.293 10.782
------------------------------------------
ary3 (1000000)  9.567 10.209  9.255  9.097

It also looks strange: D is quite good on fibo, but bad on ackermann; again good at array, but bad on matrix.

Walter, has you got time to look at these issues?

> why D is so bad on the follwing 4 tests in the middle? especially the hashes?

I also thought the hash's benchmark strange, and I checked it.

http://user.ecc.u-tokyo.ac.jp/~s31552/wp/d/hashbench2_en.rd.html

-------
 shinichiro.h

no@where.no wrote:
> I think object persistence is the last missing feature that keep me from jumping
> to D.  Walter, what's your roadmap for this?

I suggest you go to www.opend.org and download DLI, the old D for Linux  compiler. In the source of its Phobos, you will find pickle.d - take it and see if you can make it work with current DMD versions.

-eye

Thanks for the pointer.  I'm wondering why it's not included in the standard distribution? since both win32 and linux version are now distributed togather.

Also, I just quickly browsed the code, and find the dump(obj) function is using a linear search.  Will this be too slow, should a hash-table be used here?

/* Store the object */
void dump (Object object)
{
dumpClass (object.classinfo); /* Save the type */

/* See if a handle for it already exists */
for (int c = 0; c < objects.length; c ++)
{
if (objects [c] === object)
{
jar.uintSave (c + 1);
return;
}
}

/* Not found, output it */
objects ~= object; /* Register the object */
jar.uintSave (objects.length); /* Save the handle */
dumpCoreClass (object.classinfo, object); /* Save the object data */
}


In article <bnlvbs$2t0e$1@digitaldaemon.com>, Ilya Minkov says...
>
>no@where.no wrote:
>> I think object persistence is the last missing feature that keep me from jumping to D.  Walter, what's your roadmap for this?
>
>I suggest you go to www.opend.org and download DLI, the old D for Linux
>  compiler. In the source of its Phobos, you will find pickle.d - take
>it and see if you can make it work with current DMD versions.
>
>-eye
>

Thanks for the pointer.  I'm wondering why it's not included in the standard distribution? since both win32 and linux version are now distributed togather.

Also, I just quickly browsed the code, and find the dump(obj) function is using a linear search.  Will this be too slow, should a hash-table be used here?

/* Store the object */
void dump (Object object)
{
dumpClass (object.classinfo); /* Save the type */

/* See if a handle for it already exists */
for (int c = 0; c < objects.length; c ++)
{
if (objects [c] === object)
{
jar.uintSave (c + 1);
return;
}
}

/* Not found, output it */
objects ~= object; /* Register the object */
jar.uintSave (objects.length); /* Save the handle */
dumpCoreClass (object.classinfo, object); /* Save the object data */
}


In article <bnlvbs$2t0e$1@digitaldaemon.com>, Ilya Minkov says...
>
>no@where.no wrote:
>> I think object persistence is the last missing feature that keep me from jumping to D.  Walter, what's your roadmap for this?
>
>I suggest you go to www.opend.org and download DLI, the old D for Linux
>  compiler. In the source of its Phobos, you will find pickle.d - take
>it and see if you can make it work with current DMD versions.
>
>-eye
>

Thanks for the pointer.  I'm wondering why it's not included in the standard distribution? since both win32 and linux version are now distributed togather.

Also, I just quickly browsed the code, and find the dump(obj) function is using a linear search.  Will this be too slow, should a hash-table be used here?

/* Store the object */
void dump (Object object)
{
dumpClass (object.classinfo); /* Save the type */

/* See if a handle for it already exists */
for (int c = 0; c < objects.length; c ++)
{
if (objects [c] === object)
{
jar.uintSave (c + 1);
return;
}
}

/* Not found, output it */
objects ~= object; /* Register the object */
jar.uintSave (objects.length); /* Save the handle */
dumpCoreClass (object.classinfo, object); /* Save the object data */
}


In article <bnlvbs$2t0e$1@digitaldaemon.com>, Ilya Minkov says...
>
>no@where.no wrote:
>> I think object persistence is the last missing feature that keep me from jumping to D.  Walter, what's your roadmap for this?
>
>I suggest you go to www.opend.org and download DLI, the old D for Linux
>  compiler. In the source of its Phobos, you will find pickle.d - take
>it and see if you can make it work with current DMD versions.
>
>-eye
>

<no@where.no> wrote in message news:bnkip1$140o$1@digitaldaemon.com...
>
> I think object persistence is the last missing feature that keep me from
jumping
> to D.  Walter, what's your roadmap for this?

I don't have one for that yet, sorry.

> BTW, follow the D link from (http://www.digitalmars.com/d/index.html)
>
> http://www.functionalfuture.com/d/
>
> why D is so bad on the follwing 4 tests in the middle? especially the
hashes?

I don't know. I'll have to check it out.

> MSVC7  GCC    D      C#
> fibo (42)       6.597  6.762  5.910  8.051
> -------------------------------------------
> ackermann (12)  3.005  2.929  6.064  9.017
> hash (1000000)  1.954  2.653 14.929  3.013
> hash2 (2000)    4.440  4.059 32.162 21.041
> matrix (100000) 7.039  7.051 16.293 10.782
> ------------------------------------------
> ary3 (1000000)  9.567 10.209  9.255  9.097
>
> It also looks strange: D is quite good on fibo, but bad on ackermann; again good at array, but bad on matrix.
>
> Walter, has you got time to look at these issues?
>
>
>