27 Mar
2007
27 Mar
'07
6:17 a.m.
Hi,
For our department Wiki, I really needed automatic links for acronyms
because we have a whole bunch for them. So, after couple of days playing
with code and regular expressions I managed to create Radeox filter,
which converts acronyms to links.
My questions are:
1)
Is that the right approach to do that or I should write plug-in? Can you
write a plug-in that processes every page, or you have to call it
explicitly?
2)
What is the right place to attach it to rendering? I attached it to the
end, just before link filter, but then I had to struggle with HTML. It
will be probably easier to write something that processes pure Wiki or
abstract syntax tree, but it seems to me that there is no such place in
Wiki.
If I do the processing immediately after page is loaded, macros are not
evaluated yet, therefore text generated by macros can't be processed (I
don't know is this a big deal, though).
It seems to me that macros are generating HTML instead of Wiki, which is
not very clean. On the other hand it will be hard to write event simple
macros with formatted text (e.g. "warning") without generating HTML
directly, or introducing some kind of style tags for Wiki markup.
3)
Radeox status: What is your experienece with Radeox team? Is there
anybody still working on the project? I found some some funny stuff in
the source code, which I would like to fix. They "swallowed" exceptions
in some places, so if you mis-configure filter, you will not why it
doesn't work.
27 Mar
27 Mar
9:59 a.m.
New subject: [xwiki-dev] Right approach for filtering
On 3/27/07, Zeljko Trogrlic <zeljko_t(a)post.htnet.hr> wrote:
Hi,
For our department Wiki, I really needed automatic links for acronyms
because we have a whole bunch for them. So, after couple of days playing
with code and regular expressions I managed to create Radeox filter,
which converts acronyms to links.
When you finalize it, you can post it in the code zone on xwiki.org
My questions are:
1)
Is that the right approach to do that or I should write plug-in? Can you
write a plug-in that processes every page, or you have to call it
explicitly?
Yes, that is the right approach (for the moment). Plugins are usually more
complex, converting wiki syntax to html is the job of Radeox.
2)
What is the right place to attach it to rendering? I
attached it to the
end, just before link filter, but then I had to struggle with HTML. It
will be probably easier to write something that processes pure Wiki or
abstract syntax tree, but it seems to me that there is no such place in
Wiki.
There is no global right place, it all depends on how it is influenced or it
influences the rest of the filters. You have to analyze and test this. For
example, a bug regarding links and emphasizing syntax was fixed just by
reordering the filters.
An abstract syntax tree will be available once we switch from Radeox to
wikimodel.
If I do the processing immediately after page is loaded, macros are not
evaluated yet, therefore text generated by macros
can't be processed (I
don't know is this a big deal, though).
It seems to me that macros are generating HTML instead of Wiki, which is
not very clean. On the other hand it will be hard to write event simple
macros with formatted text (e.g. "warning") without generating HTML
directly, or introducing some kind of style tags for Wiki markup.
3)
Radeox status: What is your experienece with Radeox team? Is there
anybody still working on the project? I found some some funny stuff in
the source code, which I would like to fix. They "swallowed" exceptions
in some places, so if you mis-configure filter, you will not why it
doesn't work.
The Radeox team decided that it is perfect some time ago, although there are
some bugs, filters that should have been written differently, invalid markup
generated. Because of this, and the problems mentioned above (finding the
right ordering in order to avoid processing already generated HTML, lack of
a wiki abstract tree), we decided to move to WikiModel as soon as possible.
But this "ASAP" is not going to happen soon enough, so any patch you can
provide before XWiki 1.0 will be much appreciated.
Sergiu
--
http://purl.org/net/sergiu
11:32 a.m.
Sergiu Dumitriu <sergiu.dumitriu@...> writes:
On 3/27/07, Zeljko Trogrlic
<zeljko_t(a)post.htnet.hr>
wrote:
Hi,For our department Wiki, I really needed automatic
links for
acronymsbecause we have a whole bunch for them. So, after couple of days
playingwith code and regular expressions I managed to create Radeox filter,
which converts acronyms to links.
When you finalize it, you can post it in the code zone on xwiki.org
I think I'm done for now and I will post it ASAP. I'm not sure where, though.
Maybe under extensions? It doesn't fit into any category. Basically, it's a
simple Java class and a small configuration change.
The Radeox team decided that it is perfect some time
ago, although there are
some bugs, filters that should have been written differently, invalid markup
generated. Because of this, and the problems mentioned above (finding the right
ordering in order to avoid processing already generated HTML, lack of a wiki
abstract tree), we decided to move to WikiModel as soon as possible. But this
"ASAP" is not going to happen soon enough, so any patch you can provide before
XWiki
1.0 will be much appreciated.Sergiu--
http://purl.org/net/sergiu
OK, but where to put fix for Radeox? Do you keep the copy of their source tree,
too?
4:56 p.m.
New subject: [xwiki-dev] Re: Right approach for filtering
On Mar 27, 2007, at 11:32 AM, Zeljko Trogrlic wrote:
Is that required for your code to work or is it just something nice
to have? If it's required for your code, then you could attach the
patch to the Plugin form and also attach a modified and patched
radeox JAR.
Of course it might be better to implement something like what I
suggested in my earlier email (I'm writing this offline now so you
may have answered to that email already...)
Thanks
-Vincent
Sergiu Dumitriu <sergiu.dumitriu@...> writes:
I'd put it in the Plugin section. Can you package it as a JAR and
attach it there. Then user drop it in the WEB-INF/lib and make the
config change you'll describe on the plugin form.
On 3/27/07, Zeljko Trogrlic
<zeljko_t(a)post.htnet.hr>
wrote:
Hi,For our department Wiki, I really needed
automatic links for
acronymsbecause we have a whole bunch for them. So, after
couple of
days
playingwith code and regular expressions I managed to create Radeox
filter,
which converts acronyms to links.
When you finalize it, you can post it in the code zone on xwiki.org
I think I'm done for now and I will post it ASAP. I'm not sure
where, though.
Maybe under extensions? It doesn't fit into any category.
Basically, it's a
simple Java class and a small configuration change. The Radeox
team decided that it is perfect some time ago, although
there are
some bugs, filters that should have been written differently,
invalid markup
generated. Because of this, and the problems mentioned above
(finding the right
ordering in order to avoid processing already generated HTML, lack
of a wiki
abstract tree), we decided to move to WikiModel as soon as
possible. But this
"ASAP" is not going to happen soon enough, so any patch you can
provide before
XWiki
1.0 will be much appreciated.Sergiu--
http://purl.org/net/sergiu
OK, but where to put fix for Radeox? Do you keep the copy of their
source tree,
too?
9:43 p.m.
Vincent Massol wrote:
No, it is not critical, but it will be easier for all of us if
exceptions are actually reported ;)
To which other email are you referring to?
OK, but where
to put fix for Radeox? Do you keep the copy of their
source tree,
too?
Is that required for your code to work or is it just something nice to
have? If it's required for your code, then you could attach the patch to
the Plugin form and also attach a modified and patched radeox JAR.
Of course it might be better to implement something like what I
suggested in my earlier email (I'm writing this offline now so you may
have answered to that email already...)
9:50 p.m.
New subject: [xwiki-dev] Re: Right approach for filtering
On Mar 27, 2007, at 9:43 PM, Zeljko Trogrlic wrote:
Vincent Massol wrote:
No, it is not critical, but it will be easier for all of us if
exceptions are actually reported ;)
To which other email are you referring to?
The one where I said:
"
Interesting problem... I think there are several possibilities:
1) You intercept page save and parse the content to add links for
acronyms.
2) You do as you suggest and play with the rendering. This means that
the links will not exist in the database though
I prefer 1) because with 2) you'll probably not benefit from features
like automatic page rename. In addition 1) is easier to implement as
we have a notification API to get called when the page is saved for
example.
What algorithm will you use to detect an acronym vs a standard word?
"
Thanks
-Vincent
OK, but
where to put fix for Radeox? Do you keep the copy of
their source tree,
too?
Is that required for your code to work or is it just something
nice to have? If it's required for your code, then you could
attach the patch to the Plugin form and also attach a modified and
patched radeox JAR.
Of course it might be better to implement something like what I
suggested in my earlier email (I'm writing this offline now so you
may have answered to that email already...)
11:37 p.m.
Vincent Massol wrote:
Was it in reply to my email? Can't see it.
To which other
email are you referring to?
The one where I said: "
Interesting problem... I think there are several possibilities:
1) You intercept page save and parse the content to add links for acronyms.
2) You do as you suggest and play with the rendering. This means that
the links will not exist in the database though
I prefer 1) because with 2) you'll probably not benefit from features
like automatic page rename. In addition 1) is easier to implement as we
have a notification API to get called when the page is saved for example.
I agree that 1) is better. It is also faster, and it is step to phase
two - automatic links. Where/how to hook?
What algorithm will you use to detect an acronym vs a
standard word?
2 letters or more, starting with upper case, letters in the middle,
numbers or upper case at the end. E.g.
MS
IPv6
RADIUS
I'm writing from home so I don't have exact regex here. It took me 3
days to figure out how to skip HTML tags and existing links.
30 Mar
30 Mar
8:06 a.m.
Vincent Massol wrote:
Interesting problem... I think there are several
possibilities:
1) You intercept page save and parse the content to add links for acronyms.
2) You do as you suggest and play with the rendering. This means that
the links will not exist in the database though
I prefer 1) because with 2) you'll probably not benefit from features
like automatic page rename. In addition 1) is easier to implement as we
have a notification API to get called when the page is saved for example.
What algorithm will you use to detect an acronym vs a standard word?
2) done and stored in plugin section.
On second thought, 1) is not that important with acronyms because they
are immutable. However, if existing renaming concept tries to rename
everything at once, and if 1) is used, then there is a race condition
between automatic links and renaming. Renaming should leave old page as
redirect page for some time. Redirect page should be detectable by link
processors (using special class?), so they can detect outdated links and
replace them eventually.
9:01 a.m.
New subject: [xwiki-dev] Re: Right approach for filtering
On 3/30/07, Zeljko Trogrlic <zeljko_t(a)post.htnet.hr> wrote:
Renaming should leave old page as
redirect page for some time. Redirect page should be detectable by link
processors (using special class?), so they can detect outdated links and
replace them eventually.
Very good observation. I do agree with this. Vincent, can you take care of
this?
--
http://purl.org/net/sergiu
9:10 a.m.
New subject: [xwiki-dev] Re: Right approach for filtering
On Mar 30, 2007, at 9:01 AM, Sergiu Dumitriu wrote:
On 3/30/07, Zeljko Trogrlic <zeljko_t(a)post.htnet.hr> wrote:
Renaming should leave old page as
redirect page for some time. Redirect page should be detectable by
link
processors (using special class?), so they can detect outdated
links and
replace them eventually.
Very good observation. I do agree with this. Vincent, can you take
care of this?
I also agree with this. We need a JIRA for it. It's probably not
going to happen before 1.0 final though (unless someone submits a
full working patch for it). It requires some changes to the data
model I believe, in order to do it properly (because deleted pages
should not be visible as normal pages).
Thanks
-Vincent
11:02 a.m.
New subject: [xwiki-dev] Re: Right approach for filtering
1. We can add a new table, xwikiredirects, which holds redirect URLs.
2. We add document.getRedirects() (maybe both ways, from and to)
3. We need can change docdoesnotexist.vm to check for redirects.
On 3/30/07, Vincent Massol <vincent(a)massol.net> wrote:
On Mar 30, 2007, at 9:01 AM, Sergiu Dumitriu wrote:
On 3/30/07, Zeljko Trogrlic <zeljko_t(a)post.htnet.hr> wrote:
--
http://purl.org/net/sergiu
Renaming should leave old page as
redirect page for some time. Redirect page should be detectable by link
processors (using special class?), so they can detect outdated links and
replace them eventually.
Very good observation. I do agree with this. Vincent, can you take care of
this?
I also agree with this. We need a JIRA for it. It's probably not going to
happen before 1.0 final though (unless someone submits a full working
patch for it). It requires some changes to the data model I believe, in
order to do it properly (because deleted pages should not be visible as
normal pages).
Thanks
-Vincent
31 Mar
31 Mar
7:08 p.m.
Sergiu Dumitriu wrote:
1. We can add a new table, xwikiredirects, which holds
redirect URLs.
2. We add document.getRedirects() (maybe both ways, from and to)
3. We need can change docdoesnotexist.vm to check for redirects.
That would be great. Or, you can take slightly different approach:
Use two tables:
* "dictionary"
* "document" (this is existing table).
In "dictionary" table, store all possible names for document. It could
be used for
* renaming
* synonyms.
Same document can have multiple names, e.g. "IP" and "Internet
Protocol". Both entries in dictionary point to the same entry in documents.
If document is renamed, create new entry and add put its name in the
redirect field of the original entry. Original entry becomes deprecated,
but it is still working.
Optionally, update all old links.
Optionally, garbage collect old synonyms.
You should probably do this after 1.0
7:26 p.m.
New subject: [xwiki-dev] Re: Right approach for filtering
On 3/31/07, Zeljko Trogrlic <zeljko_t(a)post.htnet.hr> wrote:
Sergiu Dumitriu wrote:
You mean that pointing to ..../view/Doc/IP will display the
Doc/Internet_Protocol document, but without redirecting? (this is what
MediaWiki does). Otherwise, I don't see the difference with the approach
above. I did not describe the actual structure of the table, but how it can
be used.
If document is renamed, create new entry and add put its name in the
1. We can add a new table, xwikiredirects, which
holds redirect URLs.
2. We add document.getRedirects() (maybe both ways, from and to)
3. We need can change docdoesnotexist.vm to check for redirects.
That would be great. Or, you can take slightly different approach:
Use two tables:
* "dictionary"
* "document" (this is existing table).
In "dictionary" table, store all possible names for document. It could
be used for
* renaming
* synonyms.
Same document can have multiple names, e.g. "IP" and "Internet
Protocol". Both entries in dictionary point to the same entry in
documents. redirect field of the original entry. Original entry
becomes deprecated,
but it is still working.
"It is still working" means that the page content is preserved as it is at
the moment of renaming?
Optionally, update all old links.
Optionally, garbage collect old synonyms.
Not automatically, because IP should always be a synonim for Internet
Protocol (until we make a disambiguation page). But anyway, the
redirects/synonyms table should be editable from the wiki interface.
You should probably do this after 1.0
Definitely.
Sergiu
--
http://purl.org/net/sergiu
11:47 p.m.
Sergiu Dumitriu wrote:
On 3/31/07, *Zeljko Trogrlic*
<zeljko_t(a)post.htnet.hr
<mailto:zeljko_t@post.htnet.hr>> wrote:
Sergiu Dumitriu wrote:
I just described what was on my mind, I'm not sure yet is it practical
from both technical and practical point of view.
I think that MediaWiki has a main name + redirections. "Redirected from"
is shown only if you use redirection and you can actually go to the
redirect page by clicking on link.
This is actually closer to your solution, and I think it better than my
idea to keep all synonyms equal, because it will be confusing for the
user if he writes one thing and gets another. This is why "redirected
from" message is important.
1. We can add a new table, xwikiredirects, which
holds redirect URLs.
2. We add document.getRedirects() (maybe both ways, from and to)
3. We need can change docdoesnotexist.vm to check for redirects.
That would be great. Or, you can take slightly different approach:
Use two tables:
* "dictionary"
* "document" (this is existing table).
In "dictionary" table, store all possible names for document. It could
be used for
* renaming
* synonyms.
Same document can have multiple names, e.g. "IP" and "Internet
Protocol". Both entries in dictionary point to the same entry in
documents.
You mean that pointing to ..../view/Doc/IP will display the
Doc/Internet_Protocol document, but without redirecting? (this is what
MediaWiki does). Otherwise, I don't see the difference with the approach
above. I did not describe the actual structure of the table, but how it
can be used.
If document is renamed, create new entry and add put its name in the
redirect field of the original entry. Original entry becomes
deprecated,
but it is still working.
"It is still working" means that the page content is preserved as it is
at the moment of renaming?
In this approach, documents are independent of their names. "Original
entry" means "original synonym". Document stays alive as long as it has
at least one synonym attached to it.
The point was, if synonym was renamed, XWiki should somehow discourage
user from using it, but it should still work for old links.
Optionally, update all old links.
Optionally, garbage collect old synonyms.
Not automatically, because IP should always be a synonim for Internet
Protocol (until we make a disambiguation page). But anyway, the
redirects/synonyms table should be editable from the wiki interface.
Note that there are two kinds of synonyms:
* "active" snonyms
* renamed synonyms; these should be updated/garbage collected.
Example: you described IPv4 and you gave it synonyms "IP" and "Internet
Protocol". Later you noticed that you have to add IPv6 and that you have
to rename "IP" to "IPv4". After renaming, you have 3 synonyms:
* IPv4 - active
* IP - inactive, marked as replaced by IP, and pointing to the same doc
* Internet protocol - active.
All three synonyms are pointing to the same document.
Synonym will allow some interesting features, like automatic linking. We
could even provide different linking policy: e.g. if term is used more
than once on the page, mark just first occurrence to avoid too much links.
27 Mar
27 Mar
1:55 p.m.
Sergiu Dumitriu <sergiu.dumitriu@...> writes:
The Radeox team decided that it is perfect some time
ago, although there are
some bugs, filters that should have been written
differently, invalid markup
generated. Because of this, and the problems mentioned above (finding the right
ordering in order to avoid processing already generated HTML, lack of a wiki
abstract tree), we decided to move to WikiModel as soon as possible. But this
"ASAP" is not going to happen soon enough, so any patch you can provide before
XWiki
WikiModel looks cool. Do they provide extension mechanism for macros and stuff?
6596
days inactive
6600
days old
14 comments
3 participants
participants (3)
-
Sergiu Dumitriu -
Vincent Massol -
Zeljko Trogrlic