svn rev #24204: branches/plugins2/
ghudson@MIT.EDU
ghudson at MIT.EDU
Thu Jul 22 21:20:08 EDT 2010
http://src.mit.edu/fisheye/changelog/krb5/?cs=24204
Commit By: ghudson
Log Message:
Fix a typo in the plugins2 README.BRANCH file and flesh out the
"what's wrong with" section a bit.
Changed Files:
U branches/plugins2/README.BRANCH
Modified: branches/plugins2/README.BRANCH
===================================================================
--- branches/plugins2/README.BRANCH 2010-07-22 03:13:38 UTC (rev 24203)
+++ branches/plugins2/README.BRANCH 2010-07-23 01:20:07 UTC (rev 24204)
@@ -14,7 +14,7 @@
The profile schema used in this branch is:
- [preauth]
+ [plugins]
interfacename = {
# May take multiple values; only named plugins will be enabled.
enable_only = name
@@ -173,3 +173,26 @@
type are kind of haphazard, erring on the side of using typedefs.
* The Hesiod support in server_misc.c was ripped out.
+
+* The framework does not allow built-in modules to be registered for a
+ pluggable interface after the first load operation for that
+ interface. This constraint is probably fine, but if it needs to be
+ revisited, the framework's data model will need to be made a little
+ more complicated to allow it.
+
+* For brevity, vtable constructor functions are called "init"
+ functions in the code. That name may be too vague, and may fool
+ module implementors into thinking the init function is responsible
+ for more than just vtable construction (such as the initialization
+ of private module data). This may need to be renamed to something
+ more specific like "initvt".
+
+* The pwqual vtable declarations put function signatures directly into
+ the vtable structure definition, with comments describing each
+ function's contract alongside the signature. This is consistent
+ with how the existing DAL pluggable interface is declared. An
+ alternative would be to create typedefs for each function signature
+ and place the comments describing the function contract with the
+ typedefs. The vtable definition would then be very concise, with
+ only one line per method. This would be consistent with how the
+ existing preauth pluggable interfaces are declared.
More information about the cvs-krb5
mailing list