Re: [Boost-users] Spirit Porting
Hartmut Kaiser wrote:
Just forget the IBM compiler. It's not up to the task to compile any of Spirit V2 code. We talked to Michel Wong at BoostCon about this and he said that his compiler team resources are very limited and that he does not expect this to be fixed too soon.
I was also under the impression that spirit classic would be deprecated, some time in the future, and had pencilled time to port to Spirit Qi. My main motivation for porting to QI, was the performance of spirit classic is atrocious. Currently I building on AIX and use the new IBM compiler (v11.1) but If I understand correctly you are saying dont bother trying to port from classic->Qi. ? Is it possible to provide more detailed information, about what parts of spirit Qi(I assume this is V2) the IBM compiler cant cope with ? Best regards, Ta, Avi
Just forget the IBM compiler. It's not up to the task to compile any of Spirit V2 code. We talked to Michel Wong at BoostCon about this and he said that his compiler team resources are very limited and that he does not expect this to be fixed too soon.
I was also under the impression that spirit classic would be deprecated, some time in the future, and had pencilled time to port to Spirit Qi.
Too many people out there use Spirit.Classic for it to be pulled from Boost any time soon. Even more as Wave is still based on it as well.
My main motivation for porting to QI, was the performance of spirit classic is atrocious.
That's definitely true.
Currently I building on AIX and use the new IBM compiler (v11.1) but If I understand correctly you are saying dont bother trying to port from classic->Qi. ?
It's difficult for me to suggest anything as I have no experience with the IBM compiler. All I have is second hand knowledge from my talks with Michael Wong and from what I gather the Boost regression tests. I know they are working on the compiler to improve Boost compatibility, but Spirit was not on Michaels list of important Boost libraries to support.
Is it possible to provide more detailed information, about what parts of spirit Qi(I assume this is V2) the IBM compiler cant cope with ?
If you look at the tests, you'll see that all are failing for this compiler. Regards Hartmut --------------- http://boost-spirit.com
On 10/30/2010 8:09 AM, Hartmut Kaiser wrote:
Just forget the IBM compiler. It's not up to the task to compile any of Spirit V2 code. We talked to Michel Wong at BoostCon about this and he said that his compiler team resources are very limited and that he does not expect this to be fixed too soon.
I was also under the impression that spirit classic would be deprecated, some time in the future, and had pencilled time to port to Spirit Qi.
Too many people out there use Spirit.Classic for it to be pulled from Boost any time soon. Even more as Wave is still based on it as well.
My main motivation for porting to QI, was the performance of spirit classic is atrocious.
That's definitely true.
This is not necessarily true. If you avoid the parse-trees and ASTs and know the behavior of PEG well (e.g. avoid ambiguity, etc.), then even classic builds quite efficient parsers. YMMV.
Currently I building on AIX and use the new IBM compiler (v11.1) but If I understand correctly you are saying dont bother trying to port from classic->Qi. ?
It's difficult for me to suggest anything as I have no experience with the IBM compiler. All I have is second hand knowledge from my talks with Michael Wong and from what I gather the Boost regression tests. I know they are working on the compiler to improve Boost compatibility, but Spirit was not on Michaels list of important Boost libraries to support.
Is it possible to provide more detailed information, about what parts of spirit Qi(I assume this is V2) the IBM compiler cant cope with ?
If you look at the tests, you'll see that all are failing for this compiler.
It will definitely take a lot of effort to support the IBM compiler. Last time we had a chat with Michael Wong at BoostCon 2010, we requested that the infrastructure used by Spirit be given some attention (e.g. Fusion). That's the best I can say ATM. The rest is up to us. Alas, neither Hartmut nor I have access to the IBM compiler nor have any experience with it. If Spirit classic can compile on the IBM compiler, then someone must have done some work to make it so. I don't recall who that is anymore. If you, Robert, or anyone else would want to do the same for Spirit2, then that would be very welcome. I can't speak for Hartmut, but I don't have sufficient time nor resources to support the compiler. Regards, -- Joel de Guzman http://www.boostpro.com http://spirit.sf.net
Joel de Guzman wrote:
On 10/30/2010 8:09 AM, Hartmut Kaiser wrote:
Currently I building on AIX and use the new IBM compiler (v11.1) but If I understand correctly you are saying dont bother trying to port from classic->Qi. ?
It's difficult for me to suggest anything as I have no experience with the IBM compiler. All I have is second hand knowledge from my talks with Michael Wong and from what I gather the Boost regression tests. I know they are working on the compiler to improve Boost compatibility, but Spirit was not on Michaels list of important Boost libraries to support.
Is it possible to provide more detailed information, about what parts of spirit Qi(I assume this is V2) the IBM compiler cant cope with ?
If you look at the tests, you'll see that all are failing for this compiler.
It will definitely take a lot of effort to support the IBM compiler.
It should be noted that all the tests are failing at the same spot within MPL with the same error. Also, VACPP is passing all MPL tests except one. This would suggest that it's worth the effort to fix/workaround this one failure which originates in the MPL library. Of course no one can know what's over the error horizon - but a little effort would likely yield a lot more information as to what might have to be done to permit the new spirit to work with this compiler.
Last time we had a chat with Michael Wong at BoostCon 2010, we requested that the infrastructure used by Spirit be given some attention (e.g. Fusion). That's the best I can say ATM.
The fusion tests show lots of failures on VACPP - but many of them are from the same source. It's unclear how much of the xml_archive syntax (if any) depends on those parts of fusion which aren't working here.
The rest is up to us. Alas, neither Hartmut nor I have access to the IBM compiler nor have any experience with it. If Spirit classic can compile on the IBM compiler, then someone must have done some work to make it so. I don't recall who that is anymore.
I can tell you I never did anything specifically regarding spirit and the IBM compiler. From my perspective - it just worked. This was the source of my initial enthusiasm for spirit.
If you, Robert, or anyone else would want to do the same for Spirit2, then that would be very welcome.
lol - since I did nothing before, I'm happy to do the same again.
I can't speak for Hartmut, but I don't have sufficient time nor resources to support the compiler.
I"m sorry to hear that. I hope someone with this compiler can invest some effort to see how far this can go. To repeat the above, it's quite possible that a little effort would make the current spirit sufficiently usable in many cases. That would go along way in my opinion. The source of this discussion has been the result of my (and I believe other's) misunderstanding of the evolution of spirit from 1.x to 2.x . I had presumed that this was some sort of upgrade in that 2.x would do everything the 1.x package would do plus a lot more and be faster (at least at runtime) than 1.x The way the headers are arranged and the "deprecated" warning message strongly suggested that users should upgrade to the new system and that we were being notified that support for 1.x would be expiring in some (near?) future. Now it's clear that I misinterpreted all this. I think things would have been helped alot if there was a whole new library, set of directories namespace, etc. called "spirit2". Instead of a compiler warning - which users (my self included) take as indication that my code should be changed, I think it would have been more accurate to state somewhere something like. "Support for spirit 1.x is now limited - users are encouraged to use spirit2 for new applications". I think that would have helped us a lot. Robert Ramey
On 10/31/10 1:24 AM, Robert Ramey wrote:
Joel de Guzman wrote:
On 10/30/2010 8:09 AM, Hartmut Kaiser wrote:
Currently I building on AIX and use the new IBM compiler (v11.1) but If I understand correctly you are saying dont bother trying to port from classic->Qi. ?
It's difficult for me to suggest anything as I have no experience with the IBM compiler. All I have is second hand knowledge from my talks with Michael Wong and from what I gather the Boost regression tests. I know they are working on the compiler to improve Boost compatibility, but Spirit was not on Michaels list of important Boost libraries to support.
Is it possible to provide more detailed information, about what parts of spirit Qi(I assume this is V2) the IBM compiler cant cope with ?
If you look at the tests, you'll see that all are failing for this compiler.
It will definitely take a lot of effort to support the IBM compiler.
It should be noted that all the tests are failing at the same spot within MPL with the same error. Also, VACPP is passing all MPL tests except one.
Cool. So why don't you try doing it yourself?
This would suggest that it's worth the effort to fix/workaround this one failure which originates in the MPL library.
Sure! Try it!
Of course no one can know what's over the error horizon - but a little effort would likely yield a lot more information as to what might have to be done to permit the new spirit to work with this compiler.
Last time we had a chat with Michael Wong at BoostCon 2010, we requested that the infrastructure used by Spirit be given some attention (e.g. Fusion). That's the best I can say ATM.
The fusion tests show lots of failures on VACPP - but many of them are from the same source. It's unclear how much of the xml_archive syntax (if any) depends on those parts of fusion which aren't working here.
The rest is up to us. Alas, neither Hartmut nor I have access to the IBM compiler nor have any experience with it. If Spirit classic can compile on the IBM compiler, then someone must have done some work to make it so. I don't recall who that is anymore.
I can tell you I never did anything specifically regarding spirit and the IBM compiler. From my perspective - it just worked. This was the source of my initial enthusiasm for spirit.
If you, Robert, or anyone else would want to do the same for Spirit2, then that would be very welcome.
lol - since I did nothing before, I'm happy to do the same again.
Reread that again before trying to sound cute. I never said it was you who worked on the IBM compiler. Spirit classic worked on the IBM compiler because someone made it work and I should give credit to the unknown hero. Of course it could also be the IBM guys who made it work. I don't know. All I know it didn't happen automatically, especially with a broken compiler.
I can't speak for Hartmut, but I don't have sufficient time nor resources to support the compiler.
I"m sorry to hear that. I hope someone with this compiler can invest some effort to see how far this can go. To repeat the above, it's quite possible that a little effort would make the current spirit sufficiently usable in many cases. That would go along way in my opinion.
The source of this discussion has been the result of my (and I believe other's) misunderstanding of the evolution of spirit from 1.x to 2.x . I had presumed that this was some sort of upgrade in that 2.x would do everything the 1.x package would do plus a lot more and be faster (at least at runtime) than 1.x The way the headers are arranged and the "deprecated" warning message strongly suggested that users should upgrade to the new system and that we were being notified that support for 1.x would be expiring in some (near?) future. Now it's clear that I misinterpreted all this. I think things would
Here's one of the messages: warning "This header is deprecated. Please use: boost/spirit/include/classic_core.hpp" How unclear is that? It has been there for 3 years now!
have been helped alot if there was a whole new library, set of directories namespace, etc. called "spirit2". Instead of a compiler warning - which users (my self included) take as indication that my code should be changed, I think it would have been more accurate to state somewhere something like. "Support for spirit 1.x is now limited - users are encouraged to use spirit2 for new applications". I think that would have helped us a lot.
I wouldn't go into this unproductive discussion again. Let me tell you this: stop pointing blame! It's totally nauseating. Stop pretending that you are a monopoly of what is right and correct. If you really care about the IBM platform, then work on it. Regards, -- Joel de Guzman http://www.boostpro.com http://spirit.sf.net
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Regarding IBM's compiler, I tried to obtain a testing version/demo. The vague link to trial/demo stuff on their website seems to be broken (I tried it in IE 8 on Windows and w3m on Linux. I lack the supercomputing power needed to run Firefox). Avi, I'd love to try to help support you here, but I can't do much without at least access to a trial version of the compiler. I've sent an email to IBM asking for the production version of their compiler, but I doubt that will end well. - -- Bryce Lelbach aka wash http://groups.google.com/group/ariel_devel -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.9 (GNU/Linux) iEYEARECAAYFAkzOPrwACgkQO/fqqIuE2t4vnQCfdXo+pzVg48TGMpv00RJEXGVI RlUAoPQjz5+vTMBQMv2/8Bm5K8QhPBuI =+IK5 -----END PGP SIGNATURE-----
participants (5)
-
Avi Bahra -
Bryce Lelbach -
Hartmut Kaiser -
Joel de Guzman -
Robert Ramey