tsearch filenames unlikes special symbols and numbers

I just tried on CVS HEAD and seems something is broken

postgres=# CREATE TEXT SEARCH DICTIONARY ru_ispell (
TEMPLATE = ispell,
DictFile = russian-utf8.dict,
AffFile = russian-utf8.aff,
StopWords = russian
);
ERROR: syntax error at or near "-"
LINE 3: DictFile = russian-utf8.dict,

postgres=# CREATE TEXT SEARCH DICTIONARY ru_ispell (
TEMPLATE = ispell,
DictFile = 'russian-utf8.dict',
AffFile = 'russian-utf8.aff',
StopWords = russian
);
ERROR: invalid text search configuration file name "russian-utf8.dict"

Honestly speaking, I have no time to follow constantly changed syntax,
but documentation
http://momjian.us/main/writings/pgsql/sgml/sql-createtsdictionary.html
doesn't make clear what's wrong.

Also, I'm wondering do we really need to show all schemas without
text search configurations defined ? Looks rather stranger.

postgres=# \dF
List of text search configurations
Schema | Name | Description
--------------------+------------+---------------------------------------
information_schema | |
pg_catalog | danish | Configuration for danish language
pg_catalog | dutch | Configuration for dutch language
pg_catalog | english | Configuration for english language
pg_catalog | finnish | Configuration for finnish language
pg_catalog | french | Configuration for french language
pg_catalog | german | Configuration for german language
pg_catalog | hungarian | Configuration for hungarian language
pg_catalog | italian | Configuration for italian language
pg_catalog | norwegian | Configuration for norwegian language
pg_catalog | portuguese | Configuration for portuguese language
pg_catalog | romanian | Configuration for romanian language
pg_catalog | russian | Configuration for russian language
pg_catalog | simple | simple configuration
pg_catalog | spanish | Configuration for spanish language
pg_catalog | swedish | Configuration for swedish language
pg_catalog | turkish | Configuration for turkish language
pg_temp_1 | |
pg_toast | |
pg_toast_temp_1 | |
public | |
(21 rows)

Another problem I see are broken examples of dictionary and parser in
documentation:
http://momjian.us/main/writings/pgsql/sgml/textsearch-rule-dictionary-example.html
http://momjian.us/main/writings/pgsql/sgml/textsearch-parser-example.html

Include files in dictionary example are now in tsearch directory:

#include "tsearch/ts_locale.h"
#include "tsearch/ts_public.h"
#include "tsearch/ts_utils.h"

I didn't test parser example.

Oleg

PS. Sorry, I miss last syntax changes, but I really don't understand
parenthesis and commas usage in SQL. It's so strange.
I remember Peter raised an objections at the very beginning.

On Sun, 2 Sep 2007, Pavel Stehule wrote:

Hello
I am found small bug
postgres=# CREATE TEXT SEARCH DICTIONARY cz1(TEMPLATE = ispell,DictFile= 'cs_czutf');ERROR: invalid text search configuration file name "cs_czutf"postgres=# CREATE TEXT SEARCH DICTIONARY cz1(TEMPLATE = ispell,DictFile= 'csczutf8');ERROR: invalid text search configuration file name "csczutf8"postgres=# CREATE TEXT SEARCH DICTIONARY cz1(TEMPLATE = ispell,DictFile= "csczutf8");ERROR: invalid text search configuration file name "csczutf8"postgres=# CREATE TEXT SEARCH DICTIONARY cz1(TEMPLATE = ispell,DictFile= "cs_czutf");ERROR: invalid text search configuration file name "cs_czutf"postgres=# CREATE TEXT SEARCH DICTIONARY cz1(TEMPLATE = ispell,DictFile= "csczutf");ERROR: could not open dictionary file"/usr/local/pgsql/share/tsearch_data/csczutf.dict": nen? souborem aniadres??em
regardsPavel Stehule

Regards,
Oleg
_____________________________________________________________
Oleg Bartunov, Research Scientist, Head of AstroNet (www.astronet.ru),
Sternberg Astronomical Institute, Moscow University, Russia
Internet: oleg@sai.msu.su, http://www.sai.msu.su/~megera/
phone: +007(495)939-16-83, +007(495)939-23-83

Oleg Bartunov <oleg@sai.msu.su> writes:

postgres=# CREATE TEXT SEARCH DICTIONARY ru_ispell (
TEMPLATE = ispell,
DictFile = 'russian-utf8.dict',
AffFile = 'russian-utf8.aff',
StopWords = russian
);
ERROR: invalid text search configuration file name "russian-utf8.dict"

I made it reject all but latin letters, which is the same restriction
that's in place for timezone set filenames. That might be overly
strong, but we definitely have to forbid "." and "/" (and "\" on
Windows). Do we want to restrict it to letters, digits, underscore?
Or does it need to be weaker than that?

Also, I'm wondering do we really need to show all schemas without
text search configurations defined ? Looks rather stranger.

Um ... I don't see that; I get

regression=# \dF
List of text search configurations
Schema | Name | Description
------------+------------+---------------------------------------
pg_catalog | danish | Configuration for danish language
pg_catalog | dutch | Configuration for dutch language
pg_catalog | english | Configuration for english language
pg_catalog | finnish | Configuration for finnish language
pg_catalog | french | Configuration for french language
pg_catalog | german | Configuration for german language
pg_catalog | hungarian | Configuration for hungarian language
pg_catalog | italian | Configuration for italian language
pg_catalog | norwegian | Configuration for norwegian language
pg_catalog | portuguese | Configuration for portuguese language
pg_catalog | romanian | Configuration for romanian language
pg_catalog | russian | Configuration for russian language
pg_catalog | simple | simple configuration
pg_catalog | spanish | Configuration for spanish language
pg_catalog | swedish | Configuration for swedish language
pg_catalog | turkish | Configuration for turkish language
(16 rows)

Are you sure you're using CVS-head psql?

Another problem I see are broken examples of dictionary and parser in
documentation:
http://momjian.us/main/writings/pgsql/sgml/textsearch-rule-dictionary-example.html
http://momjian.us/main/writings/pgsql/sgml/textsearch-parser-example.html

Yeah, I wanted to discuss that with you. Code examples in sgml docs are
a bad idea: they're impossible to use as actual templates, because of
all the weird markup changes, and there's no easy way to notice if
they're broken. It would be better to remove these from the docs and
set them up as contrib modules.

regards, tom lane

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

Oleg Bartunov <oleg@sai.msu.su> writes:

postgres=# CREATE TEXT SEARCH DICTIONARY ru_ispell (
TEMPLATE = ispell,
DictFile = 'russian-utf8.dict',
AffFile = 'russian-utf8.aff',
StopWords = russian
);
ERROR: invalid text search configuration file name "russian-utf8.dict"

I made it reject all but latin letters, which is the same restriction
that's in place for timezone set filenames. That might be overly
strong, but we definitely have to forbid "." and "/" (and "\" on
Windows). Do we want to restrict it to letters, digits, underscore?
Or does it need to be weaker than that?

What's the problem with "."?

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com

Gregory Stark <stark@enterprisedb.com> writes:

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

I made it reject all but latin letters, which is the same restriction
that's in place for timezone set filenames. That might be overly
strong, but we definitely have to forbid "." and "/" (and "\" on
Windows). Do we want to restrict it to letters, digits, underscore?
Or does it need to be weaker than that?

What's the problem with "."?

../../../../etc/passwd

Possibly we could allow '.' as long as we forbade /, but the other
trouble with allowing . is that it encourages people to try to specify
the filetype suffix (as indeed Oleg was doing). I'd prefer to keep the
suffixes out of the SQL object definitions, with an eye to possibly
someday migrating all the configuration data inside the database.
There's a reasonable argument for restricting the names used for these
things in the SQL definitions to be valid SQL identifiers, so that that
will work nicely...

regards, tom lane

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

Gregory Stark <stark@enterprisedb.com> writes:

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

I made it reject all but latin letters, which is the same restriction
that's in place for timezone set filenames. That might be overly
strong, but we definitely have to forbid "." and "/" (and "\" on
Windows). Do we want to restrict it to letters, digits, underscore?
Or does it need to be weaker than that?

What's the problem with "."?

../../../../etc/passwd

Possibly we could allow '.' as long as we forbade /,

Right, traditionally the only characters forbidden in filenames in Unix are /
and nul. If we want the files to play nice in Gnome etc then we should
restrict them to ascii since we don't know what encoding the gui expects.

Actually I think in Windows \ : and . are problems (not allowed more than one
dot in dos).

There's a reasonable argument for restricting the names used for these
things in the SQL definitions to be valid SQL identifiers, so that that
will work nicely...

Ah

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com

On 9/2/07, Gregory Stark <stark@enterprisedb.com> wrote:

Right, traditionally the only characters forbidden in filenames in Unix are /
and nul. If we want the files to play nice in Gnome etc then we should
restrict them to ascii since we don't know what encoding the gui expects.

Actually I think in Windows \ : and . are problems (not allowed more than one
dot in dos).

Reserved characters in Windows filenames are < > : " / \ | ? *
DOS limitations aren't relevant on the OS versions Postgres supports.

...but I thought this was about opening existing files, not creating
them, in which case the only relevant limitation is path separators.
Any other reserved characters are going to result in no open file,
rather than a security hole.

On Mon, Sep 03, 2007 at 07:47:14AM +0100, Gregory Stark wrote:

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

Gregory Stark <stark@enterprisedb.com> writes:

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

I made it reject all but latin letters, which is the same restriction
that's in place for timezone set filenames. That might be overly
strong, but we definitely have to forbid "." and "/" (and "\" on
Windows). Do we want to restrict it to letters, digits, underscore?
Or does it need to be weaker than that?

What's the problem with "."?

../../../../etc/passwd

Possibly we could allow '.' as long as we forbade /,

Right, traditionally the only characters forbidden in filenames in Unix are /
and nul. If we want the files to play nice in Gnome etc then we should
restrict them to ascii since we don't know what encoding the gui expects.

Actually I think in Windows \ : and . are problems (not allowed more than one
dot in dos).

\ and : are problems.

. is not a problem. We don't support 16-bit windows anyway, and multiple
dots works fine on any system we support.

//Magnus

Magnus Hagander <magnus@hagander.net> writes:

On Mon, Sep 03, 2007 at 07:47:14AM +0100, Gregory Stark wrote:

Actually I think in Windows \ : and . are problems (not allowed more
than one dot in dos).

\ and : are problems.

Is : really a problem, given that the name in question will be appended
to a known directory's path?

. is not a problem. We don't support 16-bit windows anyway, and multiple
dots works fine on any system we support.

I'm not convinced that . is issue-free. On most if not all versions of Unix,
you are allowed to open a directory as a file and read the filenames it
contains. While I don't say it'd be easy to manage that through
tsearch, there's at least a potential for discovering the filenames
present in . and .. --- how much do we care about that?

regards, tom lane

On Mon, Sep 03, 2007 at 09:27:19AM -0400, Tom Lane wrote:

Magnus Hagander <magnus@hagander.net> writes:

On Mon, Sep 03, 2007 at 07:47:14AM +0100, Gregory Stark wrote:

Actually I think in Windows \ : and . are problems (not allowed more
than one dot in dos).

\ and : are problems.

Is : really a problem, given that the name in question will be appended
to a known directory's path?

Yes. It won't work - the API calls will reject it.

. is not a problem. We don't support 16-bit windows anyway, and multiple
dots works fine on any system we support.

I'm not convinced that . is issue-free. On most if not all versions of Unix,
you are allowed to open a directory as a file and read the filenames it
contains. While I don't say it'd be easy to manage that through
tsearch, there's at least a potential for discovering the filenames
present in . and .. --- how much do we care about that?

I just meant that it's not a problem on Win32 to have a file with multiple
dots in the name. There can certainly be *other* reasons for it. I don't
really see the need to have an extra dot in the filename in this particular
case, so I'd certainly be fine with restricting this one a lot more.

//Magnus

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

I'm not convinced that . is issue-free. On most if not all versions of Unix,
you are allowed to open a directory as a file and read the filenames it
contains. While I don't say it'd be easy to manage that through
tsearch, there's at least a potential for discovering the filenames
present in . and .. --- how much do we care about that?

Actually I don't think that's true any more, most file systems on most Unixen
do not allow it. However it appears it's still the case for Solaris so it's
still a good point.

I'm sure it's not true for modern versions of Linux and I thought it was false
for other modern OSes -- I'm surprised it's not for Solaris even.

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com

Gregory Stark <stark@enterprisedb.com> writes:

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

I'm not convinced that . is issue-free. On most if not all versions of Unix,
you are allowed to open a directory as a file and read the filenames it
contains. While I don't say it'd be easy to manage that through
tsearch, there's at least a potential for discovering the filenames
present in . and .. --- how much do we care about that?

Actually I don't think that's true any more, most file systems on most Unixen
do not allow it. However it appears it's still the case for Solaris so it's
still a good point.

Actually, now that I've woken up a bit more, it is not a problem as
long as the tsearch code always appends some kind of file extension
to what the user gives, such as ".dict". It'll be impossible to name
"." or ".." with that addition.

Also, Magnus says that Windows throws an error for ":" in the filename,
which means we needn't.

So the bottom line seems to be that rejecting directory separators
is sufficient to prevent any unwanted file accesses.

It might still be a good idea to restrict the names to be SQL
identifiers (ie, alphanumerics and underscores) for future-proofing,
but it wasn't clear whether anyone but me thought that was a good
argument. I'm willing to make it just be no-dir-separators.

regards, tom lane

"Tom Lane" <tgl@sss.pgh.pa.us> writes:

It might still be a good idea to restrict the names to be SQL
identifiers (ie, alphanumerics and underscores) for future-proofing,
but it wasn't clear whether anyone but me thought that was a good
argument. I'm willing to make it just be no-dir-separators.

I thought that was a good argument actually.

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com

Tom Lane wrote:

Magnus Hagander <magnus@hagander.net> writes:

On Mon, Sep 03, 2007 at 07:47:14AM +0100, Gregory Stark wrote:

Actually I think in Windows \ : and . are problems (not allowed more
than one dot in dos).

\ and : are problems.

Is : really a problem, given that the name in question will be appended
to a known directory's path?

The file name shouldn't have a ':' in it. Accessing a path with multiple
':' in it to open a file for reading should just fail normally. So yes,
there should be no problem.

. is not a problem. We don't support 16-bit windows anyway, and multiple
dots works fine on any system we support.

I'm not convinced that . is issue-free. On most if not all versions of Unix,
you are allowed to open a directory as a file and read the filenames it
contains. While I don't say it'd be easy to manage that through
tsearch, there's at least a potential for discovering the filenames
present in . and .. --- how much do we care about that?

No more than discovering the file names in any other directory without
using '.' or '..'? If it matters, check to ensure it is a regular file
before opening it?

Cheers,
mark

--
Mark Mielke <mark@mielke.cc>

Tom Lane wrote:

Also, ____ says that Windows throws an error for ":" in the filename,
which means we needn't.

Windows doesn't fail - but it can do odd things. For example, try:

C:\> echo hi >foo:bar

If one then checks the directory, one finds a "foo".

Depending on *which* API one uses, the rules may change around a bit -
but whatever the situation, as long as you prefix it with a valid path,
the ":" is not going to cause you problems.

It might still be a good idea to restrict the names to be SQL
identifiers (ie, alphanumerics and underscores) for future-proofing,
but it wasn't clear whether anyone but me thought that was a good
argument. I'm willing to make it just be no-dir-separators.

I think it is a good argument.

Cheers,
mark

--
Mark Mielke <mark@mielke.cc>

On 9/3/07, Mark Mielke <mark@mark.mielke.cc> wrote:

Tom Lane wrote:

Also, ____ says that Windows throws an error for ":" in the filename,
which means we needn't.

Windows doesn't fail - but it can do odd things. For example, try:

C:\> echo hi >foo:bar

If one then checks the directory, one finds a "foo".

: is used for naming streams and attribute types in NTFS filenames.
It's not very well-known functionality and tends to confuse people,
but I'm not aware of any situation where it'd be a problem for read
access. (Creation is not a security risk in the technical sense, but
as most administrators aren't aware of alternate data streams and the
shell does not expose them, it's effectively hidden data.)

If any of you are familiar with MacOS HFS resource forks, NTFS
basically supports an arbitrary number of named forks. A file is
collection of one or more data streams, the single unnamed stream
being default.

Moving to -docs

On Sun, Sep 02, 2007 at 06:46:11PM -0400, Tom Lane wrote:

Another problem I see are broken examples of dictionary and parser in
documentation:
http://momjian.us/main/writings/pgsql/sgml/textsearch-rule-dictionary-example.html
http://momjian.us/main/writings/pgsql/sgml/textsearch-parser-example.html

Yeah, I wanted to discuss that with you. Code examples in sgml docs are
a bad idea: they're impossible to use as actual templates, because of
all the weird markup changes, and there's no easy way to notice if
they're broken. It would be better to remove these from the docs and
set them up as contrib modules.

Couldn't we come up with some method of specifying code examples in the
docs and then having the doc build process actually run those examples
and put that into the doc build?

I wrote some code that does this back when I was thinking about writing
a book, if anyone wants to see it.
--
Decibel!, aka Jim Nasby decibel@decibel.org
EnterpriseDB http://enterprisedb.com 512.569.9461 (cell)

Trevor Talbot wrote:

On 9/3/07, Mark Mielke <mark@mark.mielke.cc> wrote:

Tom Lane wrote:

Also, ____ says that Windows throws an error for ":" in the filename,
which means we needn't.

Windows doesn't fail - but it can do odd things. For example, try:

C:\> echo hi >foo:bar

If one then checks the directory, one finds a "foo".

: is used for naming streams and attribute types in NTFS filenames.
It's not very well-known functionality and tends to confuse people,
but I'm not aware of any situation where it'd be a problem for read
access. (Creation is not a security risk in the technical sense, but
as most administrators aren't aware of alternate data streams and the
shell does not expose them, it's effectively hidden data.)

If any of you are familiar with MacOS HFS resource forks, NTFS
basically supports an arbitrary number of named forks. A file is
collection of one or more data streams, the single unnamed stream
being default.

On MacOS (prior) to OSX, : was used as a directory seperator (Paths
looked like "My Harddisk:My Folder:Somefile"). In OSX, "/" is used,
but for backwards-compatibility the Finder translates "/" in filenames
to ":". So, of you do for example "touch 'my:test'" on the shell,
you see "my/test" in the Finder.

Thats another argument for staying away from : in filenames.

greetings, Florian Pflug

Tom Lane escribi�:

Possibly we could allow '.' as long as we forbade /, but the other
trouble with allowing . is that it encourages people to try to specify
the filetype suffix (as indeed Oleg was doing). I'd prefer to keep the
suffixes out of the SQL object definitions, with an eye to possibly
someday migrating all the configuration data inside the database.
There's a reasonable argument for restricting the names used for these
things in the SQL definitions to be valid SQL identifiers, so that that
will work nicely...

Well, if we were to use SQL identifiers, we couldn't forbade anything
too much, seeing as almost anything can be used as an identifier, so
long as it is properly quoted.

But it seems to me like we could just pick an convenient subset which
doesn't make any OS too angry about it (say, reject / \ . and :), and
when we get to using actual SQL identifiers, we can enlarge the
supported char set without creating any backwards-compatibility problem.

On the other hand, this means the name has to be quoted if it would be
quoted as an SQL identifier, right?

--
Alvaro Herrera http://www.amazon.com/gp/registry/DXLWNGRJD34J
"Nunca confiar� en un traidor. Ni siquiera si el traidor lo he creado yo"
(Bar�n Vladimir Harkonnen)

Alvaro Herrera <alvherre@commandprompt.com> writes:

On the other hand, this means the name has to be quoted if it would be
quoted as an SQL identifier, right?

Something like that. I wasn't planning on rejecting uppercase letters,
though, which would be necessary if you wanted to be strict about
matching unquoted identifiers.

There seems fairly clear use-case for allowing A-Z a-z 0-9 and
underscore (while CVS head rejects 0-9 and underscore). There also seem
to be good arguments for disallowing / \ : on various platforms, which
leaves us with some other punctuation in question, as well as the whole
matter of non-ASCII characters. I'm not sure whether we want to touch
the idea of non-ASCII; comments?

regards, tom lane