nsIComponentRegistrar.
2. Converts callers of nsComponentManager::AutoRegister to use
nsIComponentRegistrar's autoRegistrar method.
3. Add nsIComponentRegistrar implmentation to nsComponentManagerImpl.
4. Rearrange nsComponentManager.cpp so that related methods are in the same
place.
5. Added a C-style function NS_GetComponentRegistrar so that getting the
registrar is easier in some places.
6. Added a nsISimpleEnumerator interface on PLDHashTableEnumeratorImpl. in
this way, the same base class can support both old style and new style
enumerations.
7. Fixed a nasty bug where unregistring factories will leave the contract id
hash with a dangling pointer. Now, when unregister is called we search the
contract id hash for entries which have the given doomned cid and remove them.
Bug 115853. r=dp@netscape.com, sr=rpotts@netscape.com
plus a bunch of other fixes and cleanups:
- Fix comment misstatements of fact in nsIFile.idl, improve style slightly.
- Fix typo in comment in nsILocalFile.idl.
- Avoid gratuitous nsCString and nsXPIDLCString copy-constructions, which
entail malloc'ing, in nsLocalFileUnix.cpp's CopyTo and GetParent methods.
But do use nsXPIDLCString instead of raw nsMemory::Alloc/Free.
- Get rid of unnecessary mLL_II2L and mLL_L2II macros, use "prlong.h" API only.
Also use the LL_* macros consistently in case a Unix lacking long long type
support wants to use this code.
* BTW, the "Date" methods should be renamed to use "Time" instead -- after all
PRTime is the type, and traditionally "time" refers to the time-number, a
scalar independent of one's position on the surface of the earth, while
"date" refers to a struct full of locale-specific information derived from
time and some "environment" variables such as DST. Can we rename these
nsIFile methods before Mozilla 0.9 / Netscape 6?
- Use CHECK_mPath consistently and first, before any assertions relating to
arguments (which logically come after the 'this' parameter CHECK_mPath is
making assertions about).
- Use nsCOMPtr for singly-inheriting implementations of XPCOM interfaces, to
avoid scary-when-scaled 0-refcnt instances from being handled (these all
got a ref via QI or equivalent soon enough, but you never know). This also
removed some naked deletes.
- Canonize all paths copied into mPath to lack trailing slashes, so we don't
need to strip trailing slashes elsewhere, in higher-frequency methods (you
set path less often than you get it or a substring of it).
- ssize_t for strlen return values.
- Since shaver used a function pointer to consolidate creat/mkdir logic, but
didn't fold the necessary close of the new fd returned by non-failing creat
into the pointed-at function, I did that.
- AppendRelativePath forbids .. as a component (bounded by / or beginning or
end of string on either side), not just in the middle of fragment (so that
foo..bar is not an illegal relative pathname -- it should not be). BTW,
what the heck is the difference between NS_ERROR_FILE_UNRECOGNIZED_PATH and
...INVALID_PATH?
- SetLeafName was overallocating the new pathname buffer by failing to subtract
the old leafname's length.
- CopyTo was failing to return an NSRESULT_FOR_ERRNO(), it just called that
macro on a line by itself -- eek! It also contained redundant if (newFD
== nsnull) {...} cleanup code, it did a useless PR_GetFileInfo call, and it
leaked FDs on read or write error.
- Implemented CopyToFollowingLinks as a forwarded call to CopyTo, Unix does
not support "copying" a symlink via normal file i/o. Should we instead
lstat in CopyTo and if a link is the source of the copy, do readlink and
then symlink?
- Fixed the readlink method (GetTarget) to null-terminate the link string in
the out parameter (readlink does not do that for you).
- Lots of little nsnull vs. NULL vs. 0, == and != applied to boolean or null
literals, white-space, indentation, bracing, comment, and sloppy code order
(e.g., declaring an initialized variable that's not used till after early
returns) fixes. Also invert some return logic so that NS_OK is the normal,
least indented, final return.
