aboutsummaryrefslogtreecommitdiffstats
path: root/README.maintainer
diff options
context:
space:
mode:
Diffstat (limited to 'README.maintainer')
-rw-r--r--README.maintainer41
1 files changed, 13 insertions, 28 deletions
diff --git a/README.maintainer b/README.maintainer
index f4021f3d..e6027447 100644
--- a/README.maintainer
+++ b/README.maintainer
@@ -21,34 +21,19 @@ Release Reminders
LDFLAGS="-fsanitize=address" \
--enable-werror --disable-shared
- * Consider running tests with latest gcc and/or valgrind. To do
- this, replace, build with debugging and without shared libraries.
- In build, create z and move each executable into z. Then create a
- script called exec-z that contains:
-
- #!/bin/sh
- exec valgrind --suppressions=/tmp/a.supp -q \
- `dirname $0`/z/`basename $0` ${1+"$@"}
-
- Symlink exec-z to each executable. /tmp/a.supp can be populated
- with suppressions for libraries, for example:
-
- {
- zlib1
- Memcheck:Cond
- fun:inflateReset2
- fun:inflateInit2_
- }
- {
- index
- Memcheck:Cond
- fun:index
- fun:expand_dynamic_string_token
- fun:_dl_map_object
- fun:map_doit
- }
-
- You can generate these by running valgrind with --gen-suppressions=yes.
+ As of gcc 6.3.0, this exposes some good things but appears to also
+ have some false positive leak reports. Valgrind is more reliable
+ but also may miss some things that this catches.
+
+ * Consider running tests with latest gcc and/or valgrind. To test
+ with valgrind:
+
+ ./configure --disable-shared
+ make -j8 -k VALGRIND=1
+ make -k check NO_REBUILD=1
+
+ This moves each binary into a subdirectory and replaces it with a
+ link to make/exec-z. See make/exec-z.
* Check all open issues in the sourceforge trackers and on github.