Mail search problem.

Mail search problem.

[Donn Terry]

It was suggested that I look up some history on an i386 floating
point problem in the mail archives.  There seems to be something amiss.

A search on "All words" for "i386 float" generates 190 matches.
Since I was told that Craig was the author, I tried:

"i386 float burley".

I get 21 hits, all but one of which are "egcs-bugs mailing list
archives by thread".  (I didn't see ANY of those in the prior
query, but it's possible I missed them.)  In any case that's
not particularly useful.

Boolean expression "rbug and craig" generates an equally entertaining
but equally unhelpful result.  (I'm looking for rbug issues, 
specifically.)

I realize that Bitrange is being kind enough to provide this
service, but they might be interested in undestanding why their
search engine doesn't seem to be doing what would be expected.
Removing the indicies from the search would help, at least.

Donn
               
-- 

===================================================
Donn Terry                  mailto:donn@interix.com
Softway Systems, Inc.        http://www.interix.com
2850 McClelland Dr, Ste. 1800   Ft.Collins CO 80525
Tel: +1-970-204-9900           Fax: +1-970-204-9951
===================================================

Re: Mail search problem.

[craig]

>It was suggested that I look up some history on an i386 floating
>point problem in the mail archives.  There seems to be something amiss.
>
>A search on "All words" for "i386 float" generates 190 matches.
>Since I was told that Craig was the author, I tried:
>
>"i386 float burley".
>
>I get 21 hits, all but one of which are "egcs-bugs mailing list
>archives by thread".  (I didn't see ANY of those in the prior
>query, but it's possible I missed them.)  In any case that's
>not particularly useful.
>
>Boolean expression "rbug and craig" generates an equally entertaining
>but equally unhelpful result.  (I'm looking for rbug issues, 
>specifically.)
>
>I realize that Bitrange is being kind enough to provide this
>service, but they might be interested in undestanding why their
>search engine doesn't seem to be doing what would be expected.
>Removing the indicies from the search would help, at least.

Haven't investigated much of what you say above, but I think perhaps
you'll get better results by looking for emails in the egcs archives
with subject headings including item like:

  FLOATING-POINT INCONSISTENCIES

(As I recall, it was all-caps.)

        tq vm, (burley)

Re: Mail search problem.

[craig]

>It was suggested that I look up some history on an i386 floating
>point problem in the mail archives.  There seems to be something amiss.
>
>A search on "All words" for "i386 float" generates 190 matches.
>Since I was told that Craig was the author, I tried:
>
>"i386 float burley".
>
>I get 21 hits, all but one of which are "egcs-bugs mailing list
>archives by thread".  (I didn't see ANY of those in the prior
>query, but it's possible I missed them.)  In any case that's
>not particularly useful.
>
>Boolean expression "rbug and craig" generates an equally entertaining
>but equally unhelpful result.  (I'm looking for rbug issues, 
>specifically.)
>
>I realize that Bitrange is being kind enough to provide this
>service, but they might be interested in undestanding why their
>search engine doesn't seem to be doing what would be expected.
>Removing the indicies from the search would help, at least.

Haven't investigated much of what you say above, but I think perhaps
you'll get better results by looking for emails in the egcs archives
with subject headings including item like:

  FLOATING-POINT INCONSISTENCIES

(As I recall, it was all-caps.)

        tq vm, (burley)

Mail search problem.

[Donn Terry]

It was suggested that I look up some history on an i386 floating
point problem in the mail archives.  There seems to be something amiss.

A search on "All words" for "i386 float" generates 190 matches.
Since I was told that Craig was the author, I tried:

"i386 float burley".

I get 21 hits, all but one of which are "egcs-bugs mailing list
archives by thread".  (I didn't see ANY of those in the prior
query, but it's possible I missed them.)  In any case that's
not particularly useful.

Boolean expression "rbug and craig" generates an equally entertaining
but equally unhelpful result.  (I'm looking for rbug issues, 
specifically.)

I realize that Bitrange is being kind enough to provide this
service, but they might be interested in undestanding why their
search engine doesn't seem to be doing what would be expected.
Removing the indicies from the search would help, at least.

Donn
               
-- 

===================================================
Donn Terry                  mailto:donn@interix.com
Softway Systems, Inc.        http://www.interix.com
2850 McClelland Dr, Ste. 1800   Ft.Collins CO 80525
Tel: +1-970-204-9900           Fax: +1-970-204-9951
===================================================

Re: Mail search problem.

[Jeffrey A Law]

  In message < Pine.BSF.4.02A.9905311950070.26336-100000@dair.pair.com >you write
:
  > installation.  Right now the one at egcs.cygnus.com lacks, like not using
  > "extra_word_characters: _" (pity since I wrote that mainly for the egcs
  > mailing lists ;-)
Yea.  I need that capability yesterday :-)  Luckily I think Jason took care
of adding support for '_' today.

jeff

Re: Mail search problem.

[Geoff Hutchison]

At the risk of going off-topic, I'll try to respond to this. I'll be glad
to answer any questions about the search software off-list (or on the htdig
lists).

>A search on "All words" for "i386 float" generates 190 matches.
>Since I was told that Craig was the author, I tried:

Personally, I would have tried x86 if i386 didn't work. The search engine
has a list of synonyms, but it's based mostly on general English text.
Compiler lingo, naturally has a different subset of synonyms. :-) Clearly a
targeted synonym file would help.

>search engine doesn't seem to be doing what would be expected.
>Removing the indicies from the search would help, at least.

While I don't know about the structure of the egcs mailing archives, this
may be easier said than done. Perhaps the best way to do this naturally is
to insert <meta name="robots" content="noindex,follow"> into the indexes.
But this doesn't help older indexes unless someone does it by hand.

I also don't know if Hans-Peter is using backlink weighting, but I usually
find that ht://Dig tries to downweight mailing list indexes when it's used.


-Geoff Hutchison
Williams Students Online
http://wso.williams.edu/

Re: Mail search problem.

[Gerald Pfeifer]

FYI: That just a few days ago Jason Molenda, has set up a htDig search
engine on the egcs site itself.

Right now this engine is usable from the various index pages of the
mailing list, but real-soon-now we'll have it active for the entire
site (and will remove the Glimpse machinery then).

Gerald
-- 
Gerald "Jerry" pfeifer@dbai.tuwien.ac.at http://www.dbai.tuwien.ac.at/~pfeifer/

Re: Mail search problem.

[Hans-Peter Nilsson]

(Slightly off-topic, mostly meta about searching the mailing lists.)

> I realize that Bitrange is being kind enough to provide this
> service, but they might be interested in undestanding why their
> search engine doesn't seem to be doing what would be expected.

As Craig and Geoff Hutchinson (the ht://Dig maintainer) pointed out, the
reason was that the words you mentioned were not in the mail you searched
for, and that no synonym lists are used, as pointed out at 
<URL: http://bitrange.com/egcs/searchlimits.html > (linked from the top of
the toplevel search form).  A (hopefully) sufficient description of the
setup is linked there too.

A boolean search for "(i386 or x86 or ix86 or intel or ia32) and float and
burley" seems to cover what you're looking for.

Please tell me if you still think something is amiss.

> Removing the indicies from the search would help, at least.
Done, thanks for pointing it out.
Indexes in general (and "favorite link" lists) cause many false hits...

And please CC me when you have comments about this search engine (as said
on <URL: http://egcs.cygnus.com/lists.html >).

BTW, Maybe the search engine at bitrange.com can be retired some time
in the coming months, now that egcs.cygnus.com has a decent ht://Dig
installation.  Right now the one at egcs.cygnus.com lacks, like not using
"extra_word_characters: _" (pity since I wrote that mainly for the egcs
mailing lists ;-) and could cut down on what's in "valid_punctuation".
It also uses gif images...  It would also be nice if the search form at
<URL: http://bitrange.com/egcs/search.html > (probably edited) could be
of use somewhere at egcs.cygnus.com.
Gerry or JSM: SYN?

(Geoff's comments about putting noindex meta in the indexes does not
apply to the one at bitrange.com (related to using an external parser),
but may apply to egcs.cygnus.com.  And backlink_factor is the default 1000).

And finally, there will be no search database updates in the coming two
weeks.  See you all at the Usenix tech conference!

brgds, H-P

Re: Mail search problem.

[David L. Nicol]

> may be easier said than done. Perhaps the best way to do this naturally is
> to insert <meta name="robots" content="noindex,follow"> into the indexes.
> But this doesn't help older indexes unless someone does it by hand.

cd index_directory
find . -type f -name 'something*matching*index*files'| xargs -i \
echo perl -e \'undef \$/;\$_=\<stdin\>\;s/\<titl/\<meta name=\"robots\"
content=\"noindex,follow\"\>\<\\/titl/i\;\
print \$_ \' \< {} \> {}.tmp \; mv {}.tmp {} | sh

Naturally one would want to do this in a scratch directory before
turning it loose on the real files, to work out any escaping/quoting
problems that I left in;

leave off the final pipe through sh until it all looks right







________________________________________________________________________
  David Nicol 816.235.1187 UMKC Network Operations david@news.umkc.edu
           "The more wailing, the better" -- David B. Luby

Re: Mail search problem.

[Hans-Peter Nilsson]

(Slightly off-topic, mostly meta about searching the mailing lists.)

> I realize that Bitrange is being kind enough to provide this
> service, but they might be interested in undestanding why their
> search engine doesn't seem to be doing what would be expected.

As Craig and Geoff Hutchinson (the ht://Dig maintainer) pointed out, the
reason was that the words you mentioned were not in the mail you searched
for, and that no synonym lists are used, as pointed out at 
<URL: http://bitrange.com/egcs/searchlimits.html > (linked from the top of
the toplevel search form).  A (hopefully) sufficient description of the
setup is linked there too.

A boolean search for "(i386 or x86 or ix86 or intel or ia32) and float and
burley" seems to cover what you're looking for.

Please tell me if you still think something is amiss.

> Removing the indicies from the search would help, at least.
Done, thanks for pointing it out.
Indexes in general (and "favorite link" lists) cause many false hits...

And please CC me when you have comments about this search engine (as said
on <URL: http://egcs.cygnus.com/lists.html >).

BTW, Maybe the search engine at bitrange.com can be retired some time
in the coming months, now that egcs.cygnus.com has a decent ht://Dig
installation.  Right now the one at egcs.cygnus.com lacks, like not using
"extra_word_characters: _" (pity since I wrote that mainly for the egcs
mailing lists ;-) and could cut down on what's in "valid_punctuation".
It also uses gif images...  It would also be nice if the search form at
<URL: http://bitrange.com/egcs/search.html > (probably edited) could be
of use somewhere at egcs.cygnus.com.
Gerry or JSM: SYN?

(Geoff's comments about putting noindex meta in the indexes does not
apply to the one at bitrange.com (related to using an external parser),
but may apply to egcs.cygnus.com.  And backlink_factor is the default 1000).

And finally, there will be no search database updates in the coming two
weeks.  See you all at the Usenix tech conference!

brgds, H-P

Re: Mail search problem.

[Gerald Pfeifer]

FYI: That just a few days ago Jason Molenda, has set up a htDig search
engine on the egcs site itself.

Right now this engine is usable from the various index pages of the
mailing list, but real-soon-now we'll have it active for the entire
site (and will remove the Glimpse machinery then).

Gerald
-- 
Gerald "Jerry" pfeifer@dbai.tuwien.ac.at http://www.dbai.tuwien.ac.at/~pfeifer/

Re: Mail search problem.

[David L. Nicol]

> may be easier said than done. Perhaps the best way to do this naturally is
> to insert <meta name="robots" content="noindex,follow"> into the indexes.
> But this doesn't help older indexes unless someone does it by hand.

cd index_directory
find . -type f -name 'something*matching*index*files'| xargs -i \
echo perl -e \'undef \$/;\$_=\<stdin\>\;s/\<titl/\<meta name=\"robots\"
content=\"noindex,follow\"\>\<\\/titl/i\;\
print \$_ \' \< {} \> {}.tmp \; mv {}.tmp {} | sh

Naturally one would want to do this in a scratch directory before
turning it loose on the real files, to work out any escaping/quoting
problems that I left in;

leave off the final pipe through sh until it all looks right







________________________________________________________________________
  David Nicol 816.235.1187 UMKC Network Operations david@news.umkc.edu
           "The more wailing, the better" -- David B. Luby

Re: Mail search problem.

[Geoff Hutchison]

At the risk of going off-topic, I'll try to respond to this. I'll be glad
to answer any questions about the search software off-list (or on the htdig
lists).

>A search on "All words" for "i386 float" generates 190 matches.
>Since I was told that Craig was the author, I tried:

Personally, I would have tried x86 if i386 didn't work. The search engine
has a list of synonyms, but it's based mostly on general English text.
Compiler lingo, naturally has a different subset of synonyms. :-) Clearly a
targeted synonym file would help.

>search engine doesn't seem to be doing what would be expected.
>Removing the indicies from the search would help, at least.

While I don't know about the structure of the egcs mailing archives, this
may be easier said than done. Perhaps the best way to do this naturally is
to insert <meta name="robots" content="noindex,follow"> into the indexes.
But this doesn't help older indexes unless someone does it by hand.

I also don't know if Hans-Peter is using backlink weighting, but I usually
find that ht://Dig tries to downweight mailing list indexes when it's used.


-Geoff Hutchison
Williams Students Online
http://wso.williams.edu/