Builds fails if stdin is redirected to empty input

[Jenselme]

Hi,

While packaging csvkit for fedora and OpenSuse, we (I and @tigerfoot) found out that tests that relying on stdin fail with this message:

[   31s] ======================================================================
[   31s] ERROR: test_no_inference (tests.test_utilities.test_csvsql.TestCSVSQL)
[   31s] ----------------------------------------------------------------------
[   31s] Traceback (most recent call last):
[   31s]   File "/home/abuild/rpmbuild/BUILD/csvkit-0.9.1/tests/test_utilities/test_csvsql.py", line 37, in test_no_inference
[   31s]     utility.main()
[   31s]   File "/home/abuild/rpmbuild/BUILD/csvkit-0.9.1/csvkit/utilities/csvsql.py", line 105, in main
[   31s]     **self.reader_kwargs
[   31s]   File "/home/abuild/rpmbuild/BUILD/csvkit-0.9.1/csvkit/table.py", line 223, in from_csv
[   31s]     headers = next(rows)
[   31s]   File "/home/abuild/rpmbuild/BUILD/csvkit-0.9.1/csvkit/py3.py", line 21, in __next__
[   31s]     return next(self.reader)
[   31s] StopIteration

From what we figured out, the problem is that you add sys.stdin to the list of input files if the script is not run on a tty. But on the packages build system, stdin is redirected to something empty. This causes the build to fail.

Below is the patch that allow us to solve this issue. Is it really needed to add sys.stdin to input files?

diff -rup a/csvkit/utilities/csvsql.py b/csvkit/utilities/csvsql.py
--- a/csvkit/utilities/csvsql.py    2015-11-19 19:37:48.000000000 +0100
+++ b/csvkit/utilities/csvsql.py    2015-11-30 18:19:56.369200131 +0100
@@ -54,14 +54,6 @@ class CSVSQL(CSVKitUtility):
         else:
             table_names = []

-        # If one or more filenames are specified, we need to add stdin ourselves (if available)
-        if sys.stdin not in self.input_files:
-            try:
-                if not sys.stdin.isatty():
-                    self.input_files.insert(0, sys.stdin)
-            except:
-                pass
-
         # Create an SQLite database in memory if no connection string is specified
         if query and not connection_string:
             connection_string = "sqlite:///:memory:"
Only in b/csvkit/utilities: csvsql.py.orig
diff -rup a/tests/test_utilities/test_csvsql.py b/tests/test_utilities/test_csvsql.py
--- a/tests/test_utilities/test_csvsql.py   2015-11-19 19:37:48.000000000 +0100
+++ b/tests/test_utilities/test_csvsql.py   2015-11-30 18:18:43.737197120 +0100
@@ -90,7 +90,8 @@ class TestCSVSQL(unittest.TestCase):

             sql = output_file.getvalue()

-            self.assertTrue('CREATE TABLE stdin' in sql)
+    # This one doesn't work on build systems koji,obs etc
+   #        self.assertTrue('CREATE TABLE stdin' in sql)
             self.assertTrue('CREATE TABLE dummy' in sql)

     def test_query(self):

Originaly reported on OpenSuse packaging. See: http://opensuse.14.x6.nabble.com/python3-test-a-strange-failure-only-during-build-td5052784.html.

[jpmckinney]

How can I reproduce this error? (without setting up an OpenSUSE build system).

[jpmckinney]

Introduced in #276

Reproducible with:

< /dev/null csvsql -v --query "select a from stdin"

[onyxfish]

Paging @jeroenjanssens: do you recall the specific reason for this block of code?

I think maybe it had something to do with using a piped file and a filesystem file in the same process, though that should also be possible using the - magic filename.

[jpmckinney]

Fixed in #510 without figuring out csvsql's logic.