[ragel-users] C++11, new narrowing rules and unsigned char on ARM

[Adrian Thurston]

Hi Jan,

Apparently, on Android ARM char is unsigned. Ragel assumes it is signed 
on all architectures. What's needed is a configure check and the 
appropriate information to make it into the descriptors in common.cc.

A few others have encountered this bug lately.

I'm currently pretty busy with Ragel 7, as soon as that's done I'm going 
to sweep for the various bugs that have been reported.

Regards,
  Adrian

On 13-09-18 10:05 AM, Jan Kundrát wrote:
> Hi Adrian,
> I'm using Ragel for parsing of e-mail headers as per RFC5322. When
> porting this project to ARM (MeeGo Harmattan, Nokia N9), I've noticed
> that it fails to build after I enable C++11 features using clang 3.3.
> Here is the error message:
>
> /home/jkt/work/prog/trojita/_build_harmattan/Rfc5322HeaderParser.generated.cpp:164:26:
> error: constant expression evaluates to -128 which cannot be narrowed to
> type 'char' [-Wc++11-narrowing]
> 39, 42, 127, 10, 9, 32, -128, -1, ^~~~
> /home/jkt/work/prog/trojita/_build_harmattan/Rfc5322HeaderParser.generated.cpp:164:26:
> note: override this message by inserting an explicit cast
> 39, 42, 127, 10, 9, 32, -128, -1, ^~~~
> static_cast<char>( )
>
> So the problem is that the signedness rules for the host (machine
> running ragel producing the .cpp file with parser) and target (the
> target platform of the C++ compiler which is producing Ragel's output)
> do not match. That's a big problem, and it isn't limited just to chars,
> actually -- because Ragel's code uses platform's native types instead of
> the portable ones, there's no guarantee that ragel's int can fit the
> data of the target's size, etc. I've solved this by patching ragel to
> use C's int<num>_t types, please see the commit at [1]. That patch fixes
> my problem.
>
> In addition, before I realized that I'm actually looking for the ragel-6
> branch, I spent some time playing with master before I finding out that
> it's some kind of a rewrite. The same bug applies there as well. Before
> I was able to get the master branch to build on my system, I had to make
> the following changes:
>
> * Fixing a build failure due to a bug in colm's headers [2]. I see
> you're upstream for that project, too, perhaps you can fix it in there
> as well.
>
> * Making sure that version.h is generated [3].
>
> After that, I simply added a rule to always treat chars as signed [4].
> That's arguably a wrong change; the code shall probably do the same
> thing as [1]. I don't have time for this now.
>
> And finally, when I tried the patched master, I found out that my ragel
> parser won't compile anymore -- please see the file at [5]. The error
> message I get is that it cannot find the "CRLF" symbol which is defined
> in the included .rl file [6]. That looks like a regression in the rewrite.
>
> Thanks for a cool software -- I hope these patches are OK and that
> you'll merge them.
>
> Cheers,
> Jan
>
> [1]
> http://repo.or.cz/w/ragel-jkt.git/commitdiff/dc238e78cd3024889b6fb2618fe5bbc20179a132
>
> [2]
> http://repo.or.cz/w/ragel-jkt.git/commitdiff/faee23876c6b5abde368355e14d786aba2300d4c
>
> [3]
> http://repo.or.cz/w/ragel-jkt.git/commitdiff/a980ec473ee66ecb6dd3cc972819c33da8d1a8d7
>
> [4]
> http://repo.or.cz/w/ragel-jkt.git/commitdiff/06fab1367f2b3d6df6d51aa2cfeb97737617fa19
>
> [5]
> http://quickgit.kde.org/?p=trojita.git&a=blob&h=efb0307c829d1c0c7939a556dd40427779221651&hb=e6dd1668fbebd3f3e676f17a5ac2acde99629ca7&f=src/Imap/Parser/Rfc5322HeaderParser.cpp
>
> [6]
> http://quickgit.kde.org/?p=trojita.git&a=blob&h=65e67a87c727714783bd793b43824795d0e94ef6&hb=e6dd1668fbebd3f3e676f17a5ac2acde99629ca7&f=src/Imap/Parser/rfc5322.rl
>
>

_______________________________________________
ragel-users mailing list
ragel-users at complang.org
http://www.complang.org/mailman/listinfo/ragel-users