diff options
| author | Kaz Kylheku <kaz@kylheku.com> | 2016-10-15 21:36:51 -0700 |
|---|---|---|
| committer | Kaz Kylheku <kaz@kylheku.com> | 2016-10-15 21:36:51 -0700 |
| commit | 969dff328ab43e3a96248fc9cce41584819b9114 (patch) | |
| tree | 58e21da914d19bf1abe5f9ec75a63acc3fd3a9b2 | |
| parent | 6a5ea25622c3dbab61b35bab87506c83206ec5fd (diff) | |
| download | txr-969dff328ab43e3a96248fc9cce41584819b9114.tar.gz txr-969dff328ab43e3a96248fc9cce41584819b9114.tar.bz2 txr-969dff328ab43e3a96248fc9cce41584819b9114.zip | |
Bugfix: unable to assign to x.y.z place.
This regression was caused by commit
957f80f "Bugfix: issue with expansion of place macros"
on Sep 7, 2016. The commit itself is sound, but exposes a
hidden problem in nearby code.
* share/txr/stdlib/place.tl (sys:pl-expand): The conditions
for terminating the loop and returning the expansion are too
weak, due to the inclusion of an incorrect test. We must not
bail when the expansion of a compound form is a compound form
with the same symbol. This is because some macros behave that
way, such as, oh: qref! The expansion of (qref a b c) is
(qref (slot a 'b) 'c): another qref form. To fully expand, we
must keep iterating until the returned form is eq to the input
form. The original macroexpand (which was replaced by
macroexpand-1 in 957f80f) hid this problem because macroexpand
doesn't use this broken termination test.
| -rw-r--r-- | share/txr/stdlib/place.tl | 5 |
1 files changed, 1 insertions, 4 deletions
diff --git a/share/txr/stdlib/place.tl b/share/txr/stdlib/place.tl index eb827fd4..e3ba6c0f 100644 --- a/share/txr/stdlib/place.tl +++ b/share/txr/stdlib/place.tl @@ -156,10 +156,7 @@ (sys:setq place (macroexpand-1 place env)) (when (or (eq place unex-place) (null place) - (and (atom place) (not (symbolp place))) - (and (consp place) - (consp unex-place) - (eq (car place) (car unex-place)))) + (and (atom place) (not (symbolp place)))) (return place)) (sys:setq unex-place place)))) |