#7583: clarify discussion of hard tab expansion in doctests.
diff --git a/Doc/library/doctest.rst b/Doc/library/doctest.rst
index 6ffd2ab..edf4a70 100644
--- a/Doc/library/doctest.rst
+++ b/Doc/library/doctest.rst
@@ -299,15 +299,8 @@
How are Docstring Examples Recognized?
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-In most cases a copy-and-paste of an interactive console session works fine, but
-doctest isn't trying to do an exact emulation of any specific Python shell. All
-hard tab characters are expanded to spaces, using 8-column tab stops. If you
-don't believe tabs should mean that, too bad: don't use hard tabs, or write
-your own :class:`DocTestParser` class.
-
-.. versionchanged:: 2.4
- Expanding tabs to spaces is new; previous versions tried to preserve hard tabs,
- with confusing results.
+In most cases a copy-and-paste of an interactive console session works fine,
+but doctest isn't trying to do an exact emulation of any specific Python shell.
::
@@ -342,6 +335,21 @@
``<BLANKLINE>`` was added; there was no way to use expected output containing
empty lines in previous versions.
+* All hard tab characters are expanded to spaces, using 8-column tab stops.
+ Tabs in output generated by the tested code are not modified. Because any
+ hard tabs in the sample output *are* expanded, this means that if the code
+ output includes hard tabs, the only way the doctest can pass is if the
+ :const:`NORMALIZE_WHITESPACE` option or directive is in effect.
+ Alternatively, the test can be rewritten to capture the output and compare it
+ to an expected value as part of the test. This handling of tabs in the
+ source was arrived at through trial and error, and has proven to be the least
+ error prone way of handling them. It is possible to use a different
+ algorithm for handling tabs by writing a custom :class:`DocTestParser` class.
+
+ .. versionchanged:: 2.4
+ Expanding tabs to spaces is new; previous versions tried to preserve hard tabs,
+ with confusing results.
+
* Output to stdout is captured, but not output to stderr (exception tracebacks
are captured via a different means).
@@ -1872,4 +1880,3 @@
.. [#] Examples containing both expected output and an exception are not supported.
Trying to guess where one ends and the other begins is too error-prone, and that
also makes for a confusing test.
-