Re: [gem5-dev] annoying rebuild of swig wrappers during NULL build
Note that Swig is about to be removed and replaced with Pybind11. Perhaps the problem will still be there, but don’t invest time if that’s not the case. Andreas On 22/04/2017, 00:55, "gem5-dev on behalf of Steve Reinhardt"wrote: >Yup, I remember getting frustrated by this and looking into it and >realizing that the problem is that the dependencies get reordered. Seemed >like the right thing would be to sort the dependencies at the right point >so that the insertion order doesn't matter, but I never figured out when >or >how to do that, so I gave up. Best wishes on finding a solution! > >Steve > > >On Fri, Apr 21, 2017 at 4:36 PM, Gabe Black wrote: > >> Hi folks. I've been looking into why the swig wrappers are all rebuilt >>even >> after a successful build when nothing has changed. I think I know why, >>but >> I'm not quite sure how to fix it. >> >> Basically, when scons scans the files looking for dependencies using the >> ClassicCPP scanner class (which has been used to build a swig scanner), >>it >> will recursively look through dependencies which are identified in an >> earlier pass. This makes sense for regular-ole-C, where that might be >>the >> only way to identify dependencies from nested includes. If it gets to a >> point where there's an include and there isn't a file for it yet >>(because >> it's a generated file probably), then it will stop recursing. Later on, >> when that file is actually generated, its dependencies will be scanned >>for, >> and will transitively become dependencies for the original file (I >>think). >> >> This leads to a problem where the files involved don't exist in the >>first >> build, but older versions of them do exist in subsequent builds. The >>first >> build will stop recursing when it hits those files and their >>dependencies >> won't be added. The later builds will be able to recurse into those >>files, >> however, and they'll add those dependencies right away. The list of >> dependencies are correct in either case (at least as far as I can tell), >> but because they're added in a different way the order is different. >> Because scons is really picky, the dependency ordering change causes >>those >> files to be "rebuilt", resulting in an identical tree. >> >> Subsequent NULL builds will have the same dependency ordering as the >>second >> build, assuming nothing else has changed, and so the annoying swig >>rebuild >> won't happen. >> >> At this point I'm sort of at a loss as to what to do, because this >>seems to >> just be something broken in how scons handles scanning for dependencies >>in >> generated files. If anybody has any ideas, please let me know. This >>isn't >> anything that urgently needs fixing, but it would be nice to knock off >>that >> rough edge. >> >> Gabe >> ___ >> gem5-dev mailing list >> gem5-dev@gem5.org >> http://m5sim.org/mailman/listinfo/gem5-dev >___ >gem5-dev mailing list >gem5-dev@gem5.org >http://m5sim.org/mailman/listinfo/gem5-dev IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you. ___ gem5-dev mailing list gem5-dev@gem5.org http://m5sim.org/mailman/listinfo/gem5-dev
Re: [gem5-dev] annoying rebuild of swig wrappers during NULL build
Yup, I remember getting frustrated by this and looking into it and realizing that the problem is that the dependencies get reordered. Seemed like the right thing would be to sort the dependencies at the right point so that the insertion order doesn't matter, but I never figured out when or how to do that, so I gave up. Best wishes on finding a solution! Steve On Fri, Apr 21, 2017 at 4:36 PM, Gabe Blackwrote: > Hi folks. I've been looking into why the swig wrappers are all rebuilt even > after a successful build when nothing has changed. I think I know why, but > I'm not quite sure how to fix it. > > Basically, when scons scans the files looking for dependencies using the > ClassicCPP scanner class (which has been used to build a swig scanner), it > will recursively look through dependencies which are identified in an > earlier pass. This makes sense for regular-ole-C, where that might be the > only way to identify dependencies from nested includes. If it gets to a > point where there's an include and there isn't a file for it yet (because > it's a generated file probably), then it will stop recursing. Later on, > when that file is actually generated, its dependencies will be scanned for, > and will transitively become dependencies for the original file (I think). > > This leads to a problem where the files involved don't exist in the first > build, but older versions of them do exist in subsequent builds. The first > build will stop recursing when it hits those files and their dependencies > won't be added. The later builds will be able to recurse into those files, > however, and they'll add those dependencies right away. The list of > dependencies are correct in either case (at least as far as I can tell), > but because they're added in a different way the order is different. > Because scons is really picky, the dependency ordering change causes those > files to be "rebuilt", resulting in an identical tree. > > Subsequent NULL builds will have the same dependency ordering as the second > build, assuming nothing else has changed, and so the annoying swig rebuild > won't happen. > > At this point I'm sort of at a loss as to what to do, because this seems to > just be something broken in how scons handles scanning for dependencies in > generated files. If anybody has any ideas, please let me know. This isn't > anything that urgently needs fixing, but it would be nice to knock off that > rough edge. > > Gabe > ___ > gem5-dev mailing list > gem5-dev@gem5.org > http://m5sim.org/mailman/listinfo/gem5-dev ___ gem5-dev mailing list gem5-dev@gem5.org http://m5sim.org/mailman/listinfo/gem5-dev
[gem5-dev] annoying rebuild of swig wrappers during NULL build
Hi folks. I've been looking into why the swig wrappers are all rebuilt even after a successful build when nothing has changed. I think I know why, but I'm not quite sure how to fix it. Basically, when scons scans the files looking for dependencies using the ClassicCPP scanner class (which has been used to build a swig scanner), it will recursively look through dependencies which are identified in an earlier pass. This makes sense for regular-ole-C, where that might be the only way to identify dependencies from nested includes. If it gets to a point where there's an include and there isn't a file for it yet (because it's a generated file probably), then it will stop recursing. Later on, when that file is actually generated, its dependencies will be scanned for, and will transitively become dependencies for the original file (I think). This leads to a problem where the files involved don't exist in the first build, but older versions of them do exist in subsequent builds. The first build will stop recursing when it hits those files and their dependencies won't be added. The later builds will be able to recurse into those files, however, and they'll add those dependencies right away. The list of dependencies are correct in either case (at least as far as I can tell), but because they're added in a different way the order is different. Because scons is really picky, the dependency ordering change causes those files to be "rebuilt", resulting in an identical tree. Subsequent NULL builds will have the same dependency ordering as the second build, assuming nothing else has changed, and so the annoying swig rebuild won't happen. At this point I'm sort of at a loss as to what to do, because this seems to just be something broken in how scons handles scanning for dependencies in generated files. If anybody has any ideas, please let me know. This isn't anything that urgently needs fixing, but it would be nice to knock off that rough edge. Gabe ___ gem5-dev mailing list gem5-dev@gem5.org http://m5sim.org/mailman/listinfo/gem5-dev