Contents of /code/trunk/maint/README

MAINTENANCE README FOR PCRE
---------------------------

The files in the "maint" directory of the PCRE source contain data, scripts, 
and programs that are used for the maintenance of PCRE, but which do not form
part of the PCRE distribution tarballs. This document describes these files and 
also contains some notes for maintainers. Its contents are:

  Files in the maint directory
  Updating to a new Unicode release
  Preparing for a PCRE release
  Making a PCRE release
  Long-term ideas (wish list)  


Files in the maint directory
----------------------------

Builducptable    A Perl script that creates the contents of the ucptable.h file
                 from two Unicode data files, which themselves are downloaded
                 from the Unicode web site. Run this script in the "maint"
                 directory.
                 
ManyConfigTests  A shell script that runs "configure, make, test" a number of
                 times with different configuration settings.
                 
Unicode.tables   The files in this directory, Scripts.txt and UnicodeData.txt, 
                 were downloaded from the Unicode web site. They contain 
                 information about Unicode characters and scripts.
                 
ucptest.c        A short C program for testing the Unicode property functions
                 in pcre_ucp_searchfuncs.c, mainly useful after rebuilding the
                 Unicode property table. Compile and run this in the "maint" 
                 directory.
                 
ucptestdata      A directory containing two files, testinput1 and testoutput1,
                 to use in conjunction with the ucptest program.
                 
utf8.c           A short, freestanding C program for converting a Unicode code
                 point into a sequence of bytes in the UTF-8 encoding, and vice
                 versa. If its argument is a hex number such as 0x1234, it
                 outputs a list of the equivalent UTF-8 bytes. If its argument
                 is sequence of concatenated UTF-8 bytes (e.g. e188b4) it
                 treats them as a UTF-8 character and outputs the equivalent
                 code point in hex.
                

Updating to a new Unicode release
---------------------------------

When there is a new release of Unicode, the files in Unicode.tables must be 
refreshed from the web site, and the Buildupctable script can then be run to 
generate a new version of ucptable.h. The ucptest program can be used to check 
that the resulting table works properly, using the data files in ucptestdata to 
check a number of test characters. 


Preparing for a PCRE release
----------------------------

This section contains a checklist of things that I consult before building a
distribution for a new release.

. Ensure that the version number and version date are correct in configure.ac,
  ChangeLog, and NEWS.

. Run ./autogen.sh to ensure everything is up-to-date.

. Compile and test with many different config options, and combinations of 
  options. The maint/ManyConfigTests script now encapsulates this testing.
  
. Run perltest.pl on the test data for tests 1 and 4. The output should match
  the PCRE test output, apart from the version identification at the top. The 
  other tests are not Perl-compatible (they use various special PCRE options).

. Test with valgrind by running "RunTest valgrind". There is also "RunGrepTest
  valgrind", though that takes quite a long time.
  
. It may also useful to test with Electric Fence, though the fact that it 
  grumbles for missing free() calls can be a nuisance. (A missing free() in 
  pcretest is hardly a big problem.) To build with EF, use:
    
    LIBS='/usr/lib/libefence.a -lpthread' with ./configure.

  Then all normal runs use it to check for buffer overflow. Also run everything
  with:
  
    EF_PROTECT_BELOW=1 <whatever>  
    
  because there have been problems with lookbehinds that looked too far. 
  
. Test with the emulated memmove() function by undefining HAVE_MEMMOVE and 
  HAVE_BCOPY in config.h. You may see a number of "pcre_memmove defined but not 
  used" warnings for the modules in which there is no call to memmove(). These
  can be ignored.

. Documentation: check AUTHORS, COPYING, ChangeLog (check date), INSTALL, 
  LICENCE, NEWS (check date), NON-UNIX-USE, and README. Many of these won't 
  need changing, but over the long term things do change.
  
. Man pages: Check all man pages for \ not followed by e or f or " because
  that indicates a markup error.

. When the release is built, test it on a number of different operating 
  systems if possible, and using different compilers as well. For example,
  on Solaris it is helpful to test using Sun's cc compiler as a change from
  gcc. Adding -xarch=v9 to the cc options does a 64-bit test, but it also
  needs -S 64 for pcretest to increase the stack size for test 2.


Making a PCRE release
---------------------

Run PrepareRelease and commit the files that it changes (by removing trailing
spaces). Then run "make distcheck" to create the tarballs and the zipball.

Don't forget to update Freshmeat when the new release is out, and to tell
webmaster@pcre.org and the mailing list.


Future ideas (wish list)
------------------------

This section records a list of ideas so that they do not get forgotten. They
vary enormously in their usefulness and potential for implementation. Some are 
very sensible; some are rather wacky. Some have been on this list for years;
others are relatively new.

. Optimization

  There are always ideas for new optimizations so as to speed up pattern 
  matching. Most of them try to save work by recognizing a non-match without 
  having to scan all the possibilities. These are some that I've recorded:

  * /((A{0,5}){0,5}){0,5}(something complex)/ on a non-matching string is very
    slow, though Perl is fast. Can we speed up somehow? Convert to {0,125}?
    OTOH, this is pathological - the user could easily fix it. 
    
  * Turn ={4} into ==== ? (for speed). I once did an experiment, and it seems
    to have little effect, and maybe makes things worse.
  
  * "Ends with literal string" - note that a single character doesn't gain much 
    over the existing "required byte" (reqbyte) feature that just saves one
    byte.
  
  * These probably need to go in study():
  
    o Remember an initial string rather than just 1 char?
    
    o A required byte from alternatives - not just the last char, but an
      earlier one if common to all alternatives.
      
    o Minimum length of subject needed.
    
    o Friedl contains other ideas.
    
. If Perl gets to a consistent state over the settings of capturing sub-
  patterns inside repeats, see if we can match it. One example of the
  difference is the matching of /(main(O)?)+/ against mainOmain, where PCRE
  leaves $2 set. In Perl, it's unset. Changing this in PCRE will be very hard
  because I think it needs much more state to be remembered.
  
. Perl 6 will be a revolution. Is it a revolution too far for PCRE? 

. Unicode

  * Note that in Perl, \s matches \pZ and similarly for \d, \w and the POSIX 
    character classes. For the moment, I've chosen not to support this for 
    backward compatibility, for speed, and because it would be messy to 
    implement.
      
  * A different approach to Unicode might be to use a typedef to do everything
    in unsigned shorts instead of unsigned chars. Actually, we'd have to have a
    new typedef to distinguish data from bits of compiled pattern that are in
    bytes, I think. There would need to be conversion functions in and out. I
    don't think this is particularly trivial - and anyway, Unicode now has
    characters that need more than 16 bits, so is this at all sensible?
  
  * There has been a request for direct support of 16-bit characters and
    UTF-16. However, since Unicode is moving beyond purely 16-bit characters,
    is this worth it at all? One possible way of handling 16-bit characters
    would be to "load" them in the same way that UTF-8 characters are loaded.
    
. Allow errorptr and erroroffset to be NULL. I don't like this idea.

. Line endings:

  * Option to use NUL as a line terminator in subject strings. This could now
    be done relatively easily since the extension to support LF, CR, and CRLF.
    If this is done, a suitable option for pcregrep is also required. 
    
. Option to provide the pattern with a length instead of with a NUL terminator.
  This probably affects quite a few places in the code. 

. Catch SIGSEGV for stack overflows?  
  
. "Cut" as described in Jeffrey Friedl's book, p364: \v and \V. The definitions
  aren't yet clear enough for me. \v flushes saved states so that no
  backtracking to anything earlier can happen; \V says "no more bumpalong", but
  does it fail the current match? As described in the book, these aren't really
  "cut" as in Prolog, are they? NOTE: (a) PCRE once had "cut", but it was
  removed when atomic groups were introduced. (b) Perl 5.10 has some (*PRUNE) 
  features -- see below.

. A feature to suspend a match via a callout was once requested.

. Option to convert results into character offsets and character lengths.

. Option for pcregrep to scan only the start of a file. I am not keen - this is 
  the job of "head".
  
. A (non-Unix) user wanted pcregrep options to (a) list a file name just once, 
  preceded by a blank line, instead of adding it to every matched line, and (b)
  support --outputfile=name.
  
. Consider making UTF-8 and UCP the default for PCRE n.0 for some n > 7.

. Add a user pointer to pcre_malloc/free functions -- some option would be 
  needed to retain backward compatibility.
  
. Define a union for the results from pcre_fullinfo().

. Provide a "random access to the subject" facility so that the way in which it 
  is stored is independent of PCRE. For efficiency, it probably isn't possible 
  to switch this dynamically. It would have to be specified when PCRE was
  compiled. PCRE would then call a function every time it wanted a character.
  
. There are new (*PRUNE) facilities in Perl 5.10, some of which it might be
  relatively easy to implement.
  
. Also in Perl 5.10 are relative subroutine references (?&-1) and (?&+1) which
  I didn't know about when I added some 5.10 features for PCRE 7.0. What about
  (?(-1)... as a condition? That's an obvious extension, even if Perl 5.10 
  doesn't have it.
  
. Wild thought: the ability to compile from PCRE's internal byte code to a real
  FSM and a very fast (third) matcher to process the result. There would be
  even more restrictions than for pcre_dfa_exec(), however. This is not easy.
  
. Should pcretest have some private locale data, to avoid relying on the
  available locales for the test data, since different OS have different ideas?
  This won't be as thorough a test, but perhaps that doesn't really matter.
  
. pcregrep: add -rs for a sorted recurse? Having to store file names and sort 
  them will of course slow it down.

. Re-arrange test 2: take out the link-size dependent stuff for a separate test 
  that is run only when the link size *is* 2; leave in some non-numbered 
  debugging tests using the new /Z feature.
  
. Stan Switzer's goto replacement for longjmp, which is apparently very slow on
  OS-X. This is used when stack recursion is disabled. It would be worth doing 
  some timing tests on other OS.

. Someone suggested --disable-callout to save code space when callouts are 
  never wanted. This seems rather marginal.  
  
. Work needs doing so that the pcregrep tests work better with different 
  linebreak settings. Currently, some tests don't work when the input files
  do not have \n line endings.
  
. If the fr_FR locale isn't available for testing, try "french" instead, 
  because this may be available on Windows. It means modifying the test data, 
  however.  

Philip Hazel
Email local part: ph10
Email domain: cam.ac.uk
Last updated: 24 April 2007
1	ph10	122	MAINTENANCE README FOR PCRE
2			---------------------------
3
4	ph10	113	The files in the "maint" directory of the PCRE source contain data, scripts,
5	ph10	122	and programs that are used for the maintenance of PCRE, but which do not form
6			part of the PCRE distribution tarballs. This document describes these files and
7			also contains some notes for maintainers. Its contents are:
8	ph10	97
9	ph10	122	Files in the maint directory
10			Updating to a new Unicode release
11			Preparing for a PCRE release
12			Making a PCRE release
13			Long-term ideas (wish list)
14
15
16			Files in the maint directory
17			----------------------------
18
19	ph10	129	Builducptable A Perl script that creates the contents of the ucptable.h file
20			from two Unicode data files, which themselves are downloaded
21			from the Unicode web site. Run this script in the "maint"
22			directory.
23
24			ManyConfigTests A shell script that runs "configure, make, test" a number of
25			times with different configuration settings.
26
27			Unicode.tables The files in this directory, Scripts.txt and UnicodeData.txt,
28			were downloaded from the Unicode web site. They contain
29			information about Unicode characters and scripts.
30
31			ucptest.c A short C program for testing the Unicode property functions
32			in pcre_ucp_searchfuncs.c, mainly useful after rebuilding the
33			Unicode property table. Compile and run this in the "maint"
34			directory.
35
36			ucptestdata A directory containing two files, testinput1 and testoutput1,
37			to use in conjunction with the ucptest program.
38
39			utf8.c A short, freestanding C program for converting a Unicode code
40			point into a sequence of bytes in the UTF-8 encoding, and vice
41			versa. If its argument is a hex number such as 0x1234, it
42			outputs a list of the equivalent UTF-8 bytes. If its argument
43			is sequence of concatenated UTF-8 bytes (e.g. e188b4) it
44			treats them as a UTF-8 character and outputs the equivalent
45			code point in hex.
46	ph10	97
47
48	ph10	122	Updating to a new Unicode release
49			---------------------------------
50
51	ph10	97	When there is a new release of Unicode, the files in Unicode.tables must be
52			refreshed from the web site, and the Buildupctable script can then be run to
53			generate a new version of ucptable.h. The ucptest program can be used to check
54			that the resulting table works properly, using the data files in ucptestdata to
55			check a number of test characters.
56	ph10	122
57
58			Preparing for a PCRE release
59			----------------------------
60
61			This section contains a checklist of things that I consult before building a
62			distribution for a new release.
63
64	ph10	155	. Ensure that the version number and version date are correct in configure.ac,
65			ChangeLog, and NEWS.
66	ph10	122
67			. Run ./autogen.sh to ensure everything is up-to-date.
68
69			. Compile and test with many different config options, and combinations of
70	ph10	129	options. The maint/ManyConfigTests script now encapsulates this testing.
71	ph10	122
72			. Run perltest.pl on the test data for tests 1 and 4. The output should match
73			the PCRE test output, apart from the version identification at the top. The
74			other tests are not Perl-compatible (they use various special PCRE options).
75
76			. Test with valgrind by running "RunTest valgrind". There is also "RunGrepTest
77			valgrind", though that takes quite a long time.
78
79	ph10	155	. It may also useful to test with Electric Fence, though the fact that it
80	ph10	122	grumbles for missing free() calls can be a nuisance. (A missing free() in
81			pcretest is hardly a big problem.) To build with EF, use:
82
83			LIBS='/usr/lib/libefence.a -lpthread' with ./configure.
84
85			Then all normal runs use it to check for buffer overflow. Also run everything
86			with:
87
88			EF_PROTECT_BELOW=1 <whatever>
89
90			because there have been problems with lookbehinds that looked too far.
91
92			. Test with the emulated memmove() function by undefining HAVE_MEMMOVE and
93	ph10	155	HAVE_BCOPY in config.h. You may see a number of "pcre_memmove defined but not
94			used" warnings for the modules in which there is no call to memmove(). These
95			can be ignored.
96	ph10	122
97			. Documentation: check AUTHORS, COPYING, ChangeLog (check date), INSTALL,
98			LICENCE, NEWS (check date), NON-UNIX-USE, and README. Many of these won't
99			need changing, but over the long term things do change.
100
101			. Man pages: Check all man pages for \ not followed by e or f or " because
102			that indicates a markup error.
103
104	ph10	155	. When the release is built, test it on a number of different operating
105			systems if possible, and using different compilers as well. For example,
106			on Solaris it is helpful to test using Sun's cc compiler as a change from
107			gcc. Adding -xarch=v9 to the cc options does a 64-bit test, but it also
108			needs -S 64 for pcretest to increase the stack size for test 2.
109	ph10	122
110	ph10	155
111	ph10	122	Making a PCRE release
112			---------------------
113
114			Run PrepareRelease and commit the files that it changes (by removing trailing
115	ph10	155	spaces). Then run "make distcheck" to create the tarballs and the zipball.
116	ph10	122
117			Don't forget to update Freshmeat when the new release is out, and to tell
118			webmaster@pcre.org and the mailing list.
119
120
121			Future ideas (wish list)
122			------------------------
123
124			This section records a list of ideas so that they do not get forgotten. They
125			vary enormously in their usefulness and potential for implementation. Some are
126			very sensible; some are rather wacky. Some have been on this list for years;
127			others are relatively new.
128
129			. Optimization
130
131			There are always ideas for new optimizations so as to speed up pattern
132			matching. Most of them try to save work by recognizing a non-match without
133			having to scan all the possibilities. These are some that I've recorded:
134
135			* /((A{0,5}){0,5}){0,5}(something complex)/ on a non-matching string is very
136			slow, though Perl is fast. Can we speed up somehow? Convert to {0,125}?
137			OTOH, this is pathological - the user could easily fix it.
138
139			* Turn ={4} into ==== ? (for speed). I once did an experiment, and it seems
140			to have little effect, and maybe makes things worse.
141
142			* "Ends with literal string" - note that a single character doesn't gain much
143			over the existing "required byte" (reqbyte) feature that just saves one
144			byte.
145
146			* These probably need to go in study():
147
148			o Remember an initial string rather than just 1 char?
149
150			o A required byte from alternatives - not just the last char, but an
151			earlier one if common to all alternatives.
152
153			o Minimum length of subject needed.
154
155			o Friedl contains other ideas.
156
157			. If Perl gets to a consistent state over the settings of capturing sub-
158			patterns inside repeats, see if we can match it. One example of the
159			difference is the matching of /(main(O)?)+/ against mainOmain, where PCRE
160			leaves $2 set. In Perl, it's unset. Changing this in PCRE will be very hard
161			because I think it needs much more state to be remembered.
162
163			. Perl 6 will be a revolution. Is it a revolution too far for PCRE?
164
165			. Unicode
166
167			* Note that in Perl, \s matches \pZ and similarly for \d, \w and the POSIX
168			character classes. For the moment, I've chosen not to support this for
169			backward compatibility, for speed, and because it would be messy to
170			implement.
171
172			* A different approach to Unicode might be to use a typedef to do everything
173			in unsigned shorts instead of unsigned chars. Actually, we'd have to have a
174			new typedef to distinguish data from bits of compiled pattern that are in
175			bytes, I think. There would need to be conversion functions in and out. I
176			don't think this is particularly trivial - and anyway, Unicode now has
177			characters that need more than 16 bits, so is this at all sensible?
178
179			* There has been a request for direct support of 16-bit characters and
180			UTF-16. However, since Unicode is moving beyond purely 16-bit characters,
181			is this worth it at all? One possible way of handling 16-bit characters
182			would be to "load" them in the same way that UTF-8 characters are loaded.
183
184			. Allow errorptr and erroroffset to be NULL. I don't like this idea.
185
186			. Line endings:
187
188			* Option to use NUL as a line terminator in subject strings. This could now
189			be done relatively easily since the extension to support LF, CR, and CRLF.
190			If this is done, a suitable option for pcregrep is also required.
191
192			. Option to provide the pattern with a length instead of with a NUL terminator.
193			This probably affects quite a few places in the code.
194
195			. Catch SIGSEGV for stack overflows?
196
197			. "Cut" as described in Jeffrey Friedl's book, p364: \v and \V. The definitions
198			aren't yet clear enough for me. \v flushes saved states so that no
199			backtracking to anything earlier can happen; \V says "no more bumpalong", but
200			does it fail the current match? As described in the book, these aren't really
201			"cut" as in Prolog, are they? NOTE: (a) PCRE once had "cut", but it was
202			removed when atomic groups were introduced. (b) Perl 5.10 has some (*PRUNE)
203			features -- see below.
204
205			. A feature to suspend a match via a callout was once requested.
206
207			. Option to convert results into character offsets and character lengths.
208
209			. Option for pcregrep to scan only the start of a file. I am not keen - this is
210			the job of "head".
211
212			. A (non-Unix) user wanted pcregrep options to (a) list a file name just once,
213			preceded by a blank line, instead of adding it to every matched line, and (b)
214			support --outputfile=name.
215
216			. Consider making UTF-8 and UCP the default for PCRE n.0 for some n > 7.
217
218			. Add a user pointer to pcre_malloc/free functions -- some option would be
219			needed to retain backward compatibility.
220
221			. Define a union for the results from pcre_fullinfo().
222
223			. Provide a "random access to the subject" facility so that the way in which it
224			is stored is independent of PCRE. For efficiency, it probably isn't possible
225			to switch this dynamically. It would have to be specified when PCRE was
226			compiled. PCRE would then call a function every time it wanted a character.
227
228			. There are new (*PRUNE) facilities in Perl 5.10, some of which it might be
229			relatively easy to implement.
230
231			. Also in Perl 5.10 are relative subroutine references (?&-1) and (?&+1) which
232			I didn't know about when I added some 5.10 features for PCRE 7.0. What about
233			(?(-1)... as a condition? That's an obvious extension, even if Perl 5.10
234			doesn't have it.
235
236			. Wild thought: the ability to compile from PCRE's internal byte code to a real
237			FSM and a very fast (third) matcher to process the result. There would be
238			even more restrictions than for pcre_dfa_exec(), however. This is not easy.
239
240			. Should pcretest have some private locale data, to avoid relying on the
241			available locales for the test data, since different OS have different ideas?
242			This won't be as thorough a test, but perhaps that doesn't really matter.
243
244			. pcregrep: add -rs for a sorted recurse? Having to store file names and sort
245			them will of course slow it down.
246
247			. Re-arrange test 2: take out the link-size dependent stuff for a separate test
248			that is run only when the link size is 2; leave in some non-numbered
249			debugging tests using the new /Z feature.
250
251			. Stan Switzer's goto replacement for longjmp, which is apparently very slow on
252			OS-X. This is used when stack recursion is disabled. It would be worth doing
253			some timing tests on other OS.
254
255			. Someone suggested --disable-callout to save code space when callouts are
256			never wanted. This seems rather marginal.
257	ph10	155
258			. Work needs doing so that the pcregrep tests work better with different
259			linebreak settings. Currently, some tests don't work when the input files
260			do not have \n line endings.
261
262			. If the fr_FR locale isn't available for testing, try "french" instead,
263			because this may be available on Windows. It means modifying the test data,
264			however.
265	ph10	122
266			Philip Hazel
267			Email local part: ph10
268			Email domain: cam.ac.uk
269	ph10	155	Last updated: 24 April 2007