Hi Stefan,
I engineered the changes to be MonetDB4 and even MonetDB5 friendly, and though PF_ROX is a proof of concept branch, the concepts will at a later stage re-appear in a main branch (certainly those on a GDK level).
However, the changes in MonetDB to support it are minimal; it is not a sacrifice to have them in a HEAD branch.
I actually think we should just make that little bit of effort to port PF_ROX to the MonetDB[4] HEAD, and take it from there. Part of that step is actually the propagation of my branch changes to the HEAD (as I was assuming would happen semi-automatically).
Peter
-----Original Message----- From: Stefan Manegold [mailto:Stefan.Manegold@cwi.nl] Sent: Wednesday, October 29, 2008 10:35 AM To: monetdb-developers@lists.sourceforge.net Cc: P.Boncz@cwi.nl Subject: Re: [Monetdb-checkins] MonetDB4/src/modules/contrib malalgebra.mx, MonetDB_4-24, 1.8.2.1, 1.8.2.2
On Wed, Oct 29, 2008 at 10:27:35AM +0100, Stefan Manegold wrote:
Gents,
given (1) the inherent incompatibility between the PF_ROX branch (forked
of
the XQuery_0-24 branch of pathfinder) and the pathfinder development
trunk
and (hence) the incompatibility between the PF_ROX branch and the development trunks of MonetDB & MonetDB4
and the facts that (2) PF_ROX (at least for now) is purely for internal research purposes and (3) cahnges ("features") in PF_ROX do (e.g.) not support updates, i.e., are far from becoming "main stream",
I guess the (only?) proper solution would (have been?) to also fork
PF_ROX
branches of the MonetDB_1-24 branch (of MonetDB) and the MonetDB_4-24
branch
(of MonetDB4) to accomodate Peter's yesterday's changes.
Note though, that this would / will exclude PF_ROX-specific changes in MonetDB & MonetDB$ from being tested --- this is already the case for the pathfinder PF_ROX branch and will be the case for the MonetDB_1-24 & MonetDB_4-24 branches, too, once we start prparing for the next release
no
later than next week.
Also, changes from the PF_ROX branches of MonetDB & MonetDB4 would/will than NOT be propagated to the respective development trunks; the same will be the case for (potential0 changes to the MonetDB_1-24 & MonetDB_4-24 branches as soon as we start prparing for the next release no later than next week.
In fact, I would even suggest to not propagate Peter's yesterday's changes to the development trunk since they are "PF_ROX-specific" (at least for now).
Stefan
See the results of "Stable" testing (once they are ready later today) to judged whether testing is "desired"/"required" or not ...
Stefan
On Tue, Oct 28, 2008 at 11:36:17PM +0100, Sjoerd Mullender wrote:
On 2008-10-28 22:37, Peter Boncz wrote:
Hi Sjoerd,
The one problem that could occur is people trying to check out
MonetDB_4-24
now, compiling it and trying to use with a precompiled MonetDB_1-24,
without
updating/recompiling that one. So, I agree with you that a problem
could
occur, but I would be surprised if any of our users would really hit
this.
Because we are going to abandon this branch I was willing to let the change pass. But in general I am opposed.
I agree that in this particular case, it is unlikely that people will encounter any problems. But again, that is because we are going to abandon this branch.
With excuses to pick nits, but contrary to you I believe that API
appends
are not "just as bad" as API modifications. An API modification in
MonetDB,
would force a release in MonetDB5, for instance. An API append does
not do
that (so it is "less bad").
It *is* bad. See the disaster involving the addition of True and False in Python in a minor release. They have sworn to never make that mistake again. People expect API stability in minor releases and
adding
something to the API is *not* stability.
PF_ROX is quite different from the HEAD in the pathfinder code due to
bot
branches making different changes to the indexing scheme; afaik this
has
prevented us from keeping it in sync with the HEAD. Various API
changes (not
appends) and even buildtool changes now force it to depend on the
stable
MonetDB/MonetDB4. For PF_ROX itself this is not a problem as it has
no
outside users. However, it might occur that this scenario will repeat (PF_ROX triggering a change/addition in MonetDB/MonetDB4 that would
trickle
towards the head via MonetDB_[14]-24).
I am not talking about the PF_ROX branch. It is not a "stable" branch and you can do with it what you want. But please do consider whether
it
is possible to use the development branches of MonetDB and MonetDB4
with
PF_ROX. If you can do that, you can make needed changes to the API
there.
Peter
-----Original Message----- From: Sjoerd Mullender [mailto:sjoerd@acm.org] Sent: Tuesday, October 28, 2008 9:50 PM To: monetdb-developers@lists.sourceforge.net Cc: monetdb-checkins@lists.sourceforge.net Subject: Re: [Monetdb-checkins] MonetDB4/src/modules/contrib
malalgebra.mx,
MonetDB_4-24, 1.8.2.1, 1.8.2.2
On 2008-10-28 20:20, Peter Boncz wrote:
Hi Sjoerd,
Thanks for keeping an eye on developer behavior and effort to
educate. My
change, however, only introduces new functions in GDK and commands
in MIL;
it
is only an API append.
API appends are just as bad as any other API change. If new code depends on the extra calls, it cannot run with the older version of
the
release. That is bad. It means that the branch is *not* "stable"
anymore.
It is checked into the stable branch because afaik PF_ROX only works
with
that. But I may as often be mistaken.
I noticed it was for PF_ROX, and that does pose a problem. It seems that the PF_ROX branch should depend on the current version of
MonetDB
and not on the stable version.
Peter
This type of change is not allowed on a stable branch. Changes of
API
are *not* allowed.
Since we're going to abandon this stable branch soon, I'll let it
pass.
On 2008-10-28 18:49, Peter Boncz wrote: > Update of /cvsroot/monetdb/MonetDB4/src/modules/contrib > In directory > 23jxhf1.ch3.sourceforge.com:/tmp/cvs-serv901/src/modules/contrib > > Modified Files: > Tag: MonetDB_4-24 > malalgebra.mx > Log Message: > changes to minimize sampling overhead (mostly in XML text index) > > - introduce bandmergejoin(), and leftmergejoin()/bandmergejoin()
with a
> cutoff > > > > U malalgebra.mx > Index: malalgebra.mx >
===================================================================
> RCS file:
/cvsroot/monetdb/MonetDB4/src/modules/contrib/malalgebra.mx,v
> retrieving revision 1.8.2.1 > retrieving revision 1.8.2.2 > diff -u -d -r1.8.2.1 -r1.8.2.2 > --- malalgebra.mx 10 Oct 2008 08:42:22 -0000 1.8.2.1 > +++ malalgebra.mx 28 Oct 2008 17:49:34 -0000 1.8.2.2 > @@ -43,7 +43,6 @@ > @:leftjoin(fetch,\nCAUTION: positional matches are assumed not to
be
> out-of-bounds!!)@ > @:leftjoin(merge)@ > @:leftjoin(hash)@ > - > .COMMAND leftthetajoin ( BAT[any::1,any::2] left,
BAT[any::2,any::3]
> right, int mode) : > BAT[any::1,any::3] = CMDleftthetajoin; > "Hook directly into the 'leftthetajoin' implementation of the
join.
> @@ -58,6 +57,22 @@ > Also, for each left tuple, all matching right tuples will
appear in
> their order > of appearrance in the right BAT. This property is handy for
XQuery
> processing." > > + .COMMAND bandmergejoin ( BAT[any::1,any::2] outer,
BAT[any::2,any::3]
> inner, > + any::2 minus, any::2 plus, bit
l_in, bit
> h_in) : > + BAT[any::1,any::3] =
CMDbandmergejoin;
> + "The bandjoin algorithm, but forced to use (left) mergejoin
rather
than
> nested loop" > + > + .COMMAND bandmergejoin ( BAT[any::1,any::2] outer,
BAT[any::2,any::3]
> inner, > + any::2 minus, any::2 plus, bit
l_in, bit
> h_in, lng limit) : > + BAT[lng,BAT] =
CMDbandmergejoin_limit;
> + "A bandjoin that uses the merge algorithm and limits the result
size.
> + Result generation is cut off after this point and the > + result [lng,BAT] holds
[estimated_full_resultsize,limited_result]"
> + > + .COMMAND leftmergejoin ( BAT[any::1,any::2] left,
BAT[any::2,any::3]
> right, lng limit) : > + BAT[lng,BAT] = CMDleftmergejoin_limit; > + "leftmergejoin, but limit the result (by cutting off result
generation)
> + return a BAT[lng,BAT] containing
[full_result_estimate,limited_result]"
> @= select > .COMMAND ord_@1select ( BAT[any::1,any::2] b, any::2 low, any::2
high)
:
> @@ -119,6 +134,7 @@ > #ifndef __MALALGEBRA_H__ > #define __MALALGEBRA_H__ > #include "gdk.h" > +#include "gdk_rangejoin.h" > > /* nothing much */ > > @@ -240,6 +256,40 @@ > return (*result = BATnlthetajoin(l, r, *mode, (size_t) *
estimate)) ?
> GDK_SUCCEED : GDK_FAIL; > } > > +int > +CMDbandmergejoin(BAT **result, BAT *left, BAT *right, ptr minus,
ptr
plus,
> bit *li, bit *hi) > +{ > + return (*result = BATbandmergejoin(left, right, minus,
plus,
*li,
> *hi)) ? GDK_SUCCEED : GDK_FAIL; > +} > + > +int > +CMDbandmergejoin_limit(BAT **result, BAT *left, BAT *right, ptr
minus,
ptr
> plus, bit *li, bit *hi, lng* limit) > +{ > + size_t cutoff = *limit; > + *result = BATbandmergejoin_limit(left, right, minus,
plus, *li,
> *hi, &cutoff); > + if (*result) { > + bat bid = (*result)->batCacheid; > + *result = BATnew(TYPE_lng, TYPE_bat, 1); > + if (*result) BUNins(*result, &cutoff, &bid,
FALSE);
> + BBPunfix(bid); > + } > + return (*result)?GDK_SUCCEED:GDK_FAIL; > +} > + > +int > +CMDleftmergejoin_limit(BAT **result, BAT *left, BAT *right, lng*
limit)
> +{ > + size_t cutoff = *limit; > + *result = BATleftmergejoin_limit(left, right, cutoff,
&cutoff);
> + if (*result) { > + bat bid = (*result)->batCacheid; > + *result = BATnew(TYPE_lng, TYPE_bat, 1); > + if (*result) BUNins(*result, &cutoff, &bid,
FALSE);
> + BBPunfix(bid); > + } > + return (*result)?GDK_SUCCEED:GDK_FAIL; > +} > + > @= selectcmd > int CMDord_@1select1(BAT **result, BAT* b, ptr value) { > return (*result = BAT_select_(b, value, 0, TRUE, TRUE, @2,
FALSE,
> TRUE))?GDK_SUCCEED:GDK_FAIL; > > >
> This SF.Net email is sponsored by the Moblin Your Move Developer's
challenge
> Build the coolest Linux based applications with Moblin SDK & win
great
> prizes > Grand prize is a trip for two to an Open Source event anywhere in
the
world
> http://moblin-contest.org/redirect.php?banner_id=100&url=/ > _______________________________________________ > Monetdb-checkins mailing list > Monetdb-checkins@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/monetdb-checkins
Sjoerd Mullender
This SF.Net email is sponsored by the Moblin Your Move Developer's
challenge
Build the coolest Linux based applications with Moblin SDK & win
great
prizes
Grand prize is a trip for two to an Open Source event anywhere in
the
world
http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Monetdb-checkins mailing list Monetdb-checkins@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-checkins
This SF.Net email is sponsored by the Moblin Your Move Developer's
challenge
Build the coolest Linux based applications with Moblin SDK & win
great
prizes
Grand prize is a trip for two to an Open Source event anywhere in
the
world
http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Monetdb-checkins mailing list Monetdb-checkins@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-checkins
-- Sjoerd Mullender
This SF.Net email is sponsored by the Moblin Your Move Developer's
challenge
Build the coolest Linux based applications with Moblin SDK & win
great
prizes Grand prize is a trip for two to an Open Source event anywhere in the
world
http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Monetdb-checkins mailing list Monetdb-checkins@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-checkins
This SF.Net email is sponsored by the Moblin Your Move Developer's
challenge
Build the coolest Linux based applications with Moblin SDK & win
great prizes
Grand prize is a trip for two to an Open Source event anywhere in the
world
http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Monetdb-checkins mailing list Monetdb-checkins@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-checkins
-- Sjoerd Mullender
--
This SF.Net email is sponsored by the Moblin Your Move Developer's
challenge
Build the coolest Linux based applications with Moblin SDK & win great
prizes
Grand prize is a trip for two to an Open Source event anywhere in the
world
http://moblin-contest.org/redirect.php?banner_id=100&url=/ _______________________________________________ Monetdb-checkins mailing list Monetdb-checkins@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/monetdb-checkins
-- | Dr. Stefan Manegold | mailto:Stefan.Manegold@cwi.nl | | CWI, P.O.Box 94079 | http://www.cwi.nl/~manegold/ | | 1090 GB Amsterdam | Tel.: +31 (20) 592-4212 | | The Netherlands | Fax : +31 (20) 592-4312 |
-- | Dr. Stefan Manegold | mailto:Stefan.Manegold@cwi.nl | | CWI, P.O.Box 94079 | http://www.cwi.nl/~manegold/ | | 1090 GB Amsterdam | Tel.: +31 (20) 592-4212 | | The Netherlands | Fax : +31 (20) 592-4312 |