#general
- h
happy-kitchen-89482
10/12/2015, 5:12 PMStu, Target is another matter, right now I'm talking about SourceRoot itself. - h
happy-kitchen-89482
10/12/2015, 5:12 PMWe can debate whether the access should be via Target or not - w
witty-crayon-22786
10/12/2015, 5:13 PMI thought that's what you meant about linking it to build_graph, apologies. - h
happy-kitchen-89482
10/12/2015, 5:13 PMNo, I just thought at first blush that build_graph might make sense as a home for source_root.py, and then realized that it doesn't. - h
happy-kitchen-89482
10/12/2015, 5:13 PMBecause I don't want us to confuse "code as a DAG" and "code as a filesystem tree" any further. - h
happy-kitchen-89482
10/12/2015, 5:14 PMI guess my real question is, does anyone have an opinion on modeling languages explicitly? We've punted on it forever, but it seems like it might be time. - h
happy-kitchen-89482
10/12/2015, 5:14 PMTo create pants/lang, even if it's skeletal initially. - w
witty-crayon-22786
10/12/2015, 5:15 PMi think a SourceRoot subsystem would only be a dependency of a
package if said package were to be createdlang - w
witty-crayon-22786
10/12/2015, 5:15 PMi don't think there is anything `lang`y about SourceRoots... they apply to resources as well, for example - a
ancient-stone-50795
10/12/2015, 5:18 PMThey are also needed for idl files like proto files which span multiple language (unless you consider protobufs a separate language) - b
bored-art-40741
10/12/2015, 5:18 PMre: source roots validating target types: That's a very easy mechanism to enforce with custom rules. I think it's totally orthogonal to source roots and doesn't belong in the same code. - b
bored-art-40741
10/12/2015, 5:19 PMSource roots are entirely about drawing a line in the filesystem tree to associate package structure with filesystem structure - h
happy-kitchen-89482
10/12/2015, 5:24 PMWell, maybe "language" is the wrong word - h
happy-kitchen-89482
10/12/2015, 5:24 PMsomething like "domain" - h
happy-kitchen-89482
10/12/2015, 5:24 PMwe created all those "is_jvm", "is_codegen" methods on Target, we were obviously grasping at some sort of concept. - h
happy-kitchen-89482
10/12/2015, 5:25 PMand if you're generating lang1 and lang2 from your codegen, it's reasonable to consider that sourceroot as associated with both languages. - h
happy-kitchen-89482
10/12/2015, 5:26 PMIn the new engine we're not going to be synthesizing targets, that was always an implementation hack. - h
happy-kitchen-89482
10/12/2015, 5:29 PMand resources might be a "language" (or "source type" or whatever we call it) in some sense - h
happy-kitchen-89482
10/12/2015, 5:30 PMWell, I won't worry about it too much for now - w
witty-crayon-22786
10/12/2015, 5:31 PMyea, agreed to some extent. it's possible they could be part of an optional mixin on language subsystems or something - w
witty-crayon-22786
10/12/2015, 5:31 PMbut on the other hand, making "languages" thin seems beneficial - h
happy-kitchen-89482
10/12/2015, 5:34 PMYeah, I was thinking of something lightweight. But really for now I just need a sensible namespace to put the replacement of source_root.py in. - h
happy-kitchen-89482
10/12/2015, 5:35 PM
?pants/lang
?pants/source - h
happy-kitchen-89482
10/12/2015, 5:36 PMI'll go with
, as it's a slightly more general concept thanpants/source
, unless anyone objects?lang - h
happy-kitchen-89482
10/12/2015, 5:36 PMThat can be where all our modeling of code-as-source (as opposed to code-as-DAG) lives. - w
witty-crayon-22786
10/12/2015, 5:53 PMworks for me - l
limited-country-20626
10/12/2015, 7:11 PMSo…@stuhood and maybe @rough-minister-58256: we are sometimes hitting stalled builds. Similar to some of the other stalls, memory is thrashing and the CPU activity drops to basically zero. This happens sometime right before the end of really big graphs. Our guess is that a thread is dying due to memory constraints and the nailgun socket closes. Python then sees empty bytes and considers that a protocol error. The failure looks like: - l
limited-country-20626
10/12/2015, 7:11 PMcompile(src/jvm/com/foursquare/path:target) failed: Problem launching via NailgunClient(host=u'127.0.0.1', port=62522, workdir='/Users/mateo/dev/foursquare.web') command org.pantsbuild.zinc.Main -log-level info -analysis-cache /Users/mateo/dev/foursquare.web/.pants.d/compile/jvm/zinc/analysis_tmpdir/src.jvm.com.foursquareXXXX MASSIVE_CLASSPATH XXXX -S-encoding -SUTF-8 -S-deprecation -S-unchecked -S-feature -S-Xfatal-warnings -S-g:vars -S-language:existentials -S-language:implicitConversions -S-language:reflectiveCalls -S-language:postfixOps -S-language:higherKinds -C-source -C1.6 -C-target -C1.6 -C-source -C1.7 -C-target -C1.7 src/jvm/com/foursquare/apiserver/snippet/tools/twitter/TwitterLink.scala: Problem in nailgun protocol with nailgun server at 127.0.0.1:62522: TruncatedHeaderError(u'While reading chunk for payload length and command, socket.recv returned no bytes (client shut down). Accumulated buffer was:\n\n’,) - l
limited-country-20626
10/12/2015, 7:12 PMwhich finally triggers after a stalled build of 20-whenever minutes. - l
limited-country-20626
10/12/2015, 7:13 PMpatrick also found it interesting that it was not gc’ing and that possibly the analysis cache needs a gc.