On 12.11.2012 14:02, jan iversen wrote:
hi
see below please
On 12 November 2012 09:30, Andre Fischer <awf....@gmail.com> wrote:
On 11.11.2012 22:06, jan iversen wrote:
Hi.
I would like to hear your opinion on the following problem:
The l10n tools uses flex, which is quite nice, however I have found two
problems with flex:
- the .l sources defines YYLMAX, but the generated .c source declares
yytext[YYLMAX] BEFORE the definition code it sucked in. I have verified
this with the flex sources, nothing to do about it. the flex generated c
sources start with flex things (including yytext[YYLMAX]) and then the
first part of the .l file is included. The manual states to use YYLMAX to
change the yytext buffer size, but with the current flex sources it does
not work.
That is a good way, especially since I can make my own targets (yy.c ->
.o) in the makefile. I will do that.
But it does not solve the name issue (yytext), I will make a bug report on
flex, and mail one I know that was involved in flex.
Sorry, I did skipped the second problem when I read your mail the first
time.
I just made an experiment with l10ntools/source/srclex.l. When I add
the line
%option prefix="bla"
before the first %% then yytext is replaced but yytext_ptr is not.
Besides yytext is not really replaced but flex adds a
#define yytext blatext
so that yytext, while not used in the rest of src_yy.c, it will be a
symbol in the object file.
That will likely cause problems when linking two lexers into the same
library/executable, which usually the reason for the renaming in the
first place.
Telling flex to produce C++ output might help (replace option -l with
--c++) but that would require additional work (at least you have to
change the compilation rules in makefile.mk to compile C++ instead of
plain C).
I am sorry to be of no help.
-Andre
Jan
- the .l sources defines a '%option prefix="genXrm_"`which according to
the
manual should replace all yy in the generated code, however it does not
work for yytext and yytext_ptr, because the are instanciated before the
code is included.
The leads to two problems:
- YYLMAX not being used, it a performance degradation on all platforms
(but
no real problem).
- yytext not being changed, is in Ubuntu (linux) not a problem, but in
windows I had to use a new swich (/forcemultiple), and I have no clue
about
mac.
I see different solutions:
1) I edit the generated .c file, and make a thick note, that if the .l
file
is translated, that has to be changed.
2) I create a sed that takes the output of flex and does the trix, this
requires changing the details of the make files.
3) I report the bug to the flex people with a patch, and we wait until a
new version is ready.
I am for version 1), but I would like to hear opinions ?
I found an answer via Google on another Apache mailing list (Apache
Thrift):
http://markmail.org/message/**hgphhaciu6nxwtk7#query:+page:**
1+mid:hgphhaciu6nxwtk7+state:**results<http://markmail.org/message/hgphhaciu6nxwtk7#query:+page:1+mid:hgphhaciu6nxwtk7+state:results>
In short:
Add "CPPFLAGS='-DYYLMAX=whatever'" to the compiler invocation (not the
call to flex that converts %lex.l to %_yy.c)
However, if I look at l10ntools/source/makefile.mk, then I don't see
an explicit compile rule for the %_yy.c files. You may have to add one.
Oh, and I think option 3) would be good to do also, without the waiting
part.
-Andre
