[SPARK-1134] Fix and document passing of arguments to IPython

Diana Carroll · mateiz · commit a599e43d6e09 · 2014-04-03T15:48:42.000-07:00
This is based on @dianacarroll's previous pull request #227, and @JoshRosen's comments on #38. Since we do want to allow passing arguments to IPython, this does the following: * It documents that IPython can't be used with standalone jobs for now. (Later versions of IPython will deal with PYTHONSTARTUP properly and enable this, see ipython/ipython#5226, but no released version has that fix.) * If you run `pyspark` with `IPYTHON=1`, it passes your command-line arguments to it. This way you can do stuff like `IPYTHON=1 bin/pyspark notebook`. * The old `IPYTHON_OPTS` remains, but I've removed it from the documentation. This is in case people read an old tutorial that uses it. This is not a perfect solution and I'd also be okay with keeping things as they are today (ignoring `$@` for IPython and using IPYTHON_OPTS), and only doing the doc change. With this change though, when IPython fixes ipython/ipython#5226, people will immediately be able to do `IPYTHON=1 bin/pyspark myscript.py` to run a standalone script and get all the benefits of running scripts in IPython (presumably better debugging and such). Without it, there will be no way to run scripts in IPython. @JoshRosen you should probably take the final call on this. Author: Diana Carroll <dcarroll@cloudera.com> Closes #294 from mateiz/spark-1134 and squashes the following commits: 747bb13 [Diana Carroll] SPARK-1134 bug with ipython prevents non-interactive use with spark; only call ipython if no command line arguments were supplied
diff --git a/bin/pyspark b/bin/pyspark
@@ -55,7 +55,8 @@ if [ -n "$IPYTHON_OPTS" ]; then
   IPYTHON=1
 fi
 
-if [[ "$IPYTHON" = "1" ]] ; then
+# Only use ipython if no command line arguments were provided [SPARK-1134]
+if [[ "$IPYTHON" = "1" && $# = 0 ]] ; then
   exec ipython $IPYTHON_OPTS
 else
   exec "$PYSPARK_PYTHON" "$@"
