This patch:
- partially inlines GetPrefValueFromEntry() at its two call sites, to avoid
repeating the kind check;
- renames it as AssignPrefValueToDomPrefValue().
MozReview-Commit-ID: BUBk3ML6h5O
PrefTypeFlags is a class with a lot of smarts (i.e. methods). PrefHashEntry is
a class with little smarts. This is silly, because PrefTypeFlags is essentially
an internal implementation detail of PrefHashEntry.
This patch merges PrefTypeFlags into PrefHashEntry, so that PrefHashEntry has
all the smarts. This means lots of `pref->mPrefFlags.Foo()` calls become
`pref->Foo()`.
The patch also changes the representation of the type and flags within
PrefHashEntry to use bitfields, which avoids the need for a Flags type and is
much simpler than the old approach.
MozReview-Commit-ID: 4Yt9OtBzh9e
It's the user value that's sticky, not the default value. Though we typically
talk about the entire pref being sticky, so that's what this patch does.
MozReview-Commit-ID: 8THuRCTZ7uw
Because it holds 8-bit strings, which only need 1 byte alignment.
For the profile on my Linux box, in every process this reduces the size of the
arena by 1 chunk, from 120 KiB to 112 KiB.
MozReview-Commit-ID: 8ozrLIq5ZZ4
This patch renames it as gInstallingCallback, adds a comment, and reduces the
scope of the AUTO_INSTALLING_CALLBACK() objects to the minimum.
MozReview-Commit-ID: EGfnBPVFtmw
This splits the measurements into several buckets, like so:
> 718,528 B (00.40%) -- preferences
> ├──262,176 B (00.14%) ── hash-table
> ├──181,952 B (00.10%) ── callbacks
> ├──122,880 B (00.07%) ── pref-name-arena
> ├───91,872 B (00.05%) ── root-branches
> ├───38,296 B (00.02%) ── string-values
> ├───21,272 B (00.01%) ── cache-data
> └───────80 B (00.00%) ── misc
The patch also measures some things that were previously overlooked.
- String pref values. (The old code had a comment that incorrectly claimed they
were allocated out of an arena.)
- The PrefCallback objects pointed to by entries in nsPrefBranch::mObservers.
And it makes the code more like typical reporters.
- It removes the "AndOtherStuff" from Preferences' measuring method, and
measuring those global structures in
PreferenceServiceReporter::CollectReports().
- It adds `const` where appropriate.
MozReview-Commit-ID: dyNg7ldQdh
This patch also adds some Set*InAnyProcess() methods, and makes nsPrefBranch a
friend of Preferences so it can call those methods.
And it moves the thin Set*() wrapper functions to Preferences.h, alongside
SetUint().
MozReview-Commit-ID: 88HhmcTFZNc
This will allow other functions to be moved into Preferences and be marked as
`private` in subsequent patches.
The patch also renames SetPrefValue() as SetValueFromDom(), because that's a
clearer name.
MozReview-Commit-ID: CB1xmPSmac6
This is unused for now, but will be necessary for nsPrefBranch::Set*() to call
into Preferences::Set*().
The patch also renames some arguments from aPref to aPrefName, because that's a
better name.
MozReview-Commit-ID: 2OPB7CHOgpw
This is nicer than a bool for tracking the Default vs. User distinction, and it
replaces the Preferences.cpp-only WhichValue type.
MozReview-Commit-ID: 8CrdDN2vBJQ
libpref only allows pref modifications in the parent process. This patch
tightens up the checking, which is a bit inconsistent.
- It removes ENSURE_MAIN_PROCESS_WITH_WARNING, which does NS_WARNING on
failure, and replaces its uses with ENSURE_MAIN_PROCESS, which does NS_ERROR
on failure. This required adding an XRE_IsParentProcess() check to one place
in editor/.
- It converts XRE_IsContentProcess() tests to !XRE_IsParentProcess(), because
we now have multiple kinds of non-parent process.
- It uses ENSURE_MAIN_PROCESS to replace other checking code in a few places.
- It improves a comment in HandleDirty().
MozReview-Commit-ID: D8znQWH7ery
InitStaticMembers() is the hottest function within libpref, because it is
called from all the getters and setters. This patch tweaks it so the common
case is handled with a single comparison, instead of two. This is valid because
sPreferences is nulled at the same time that sShutdown is set.
The patch also tweaks some other conditions to take advantage of this fact.
MozReview-Commit-ID: C74fLWOw7bE
They both have two callsites, but in one of those they appear together, where
much of the code can be replaced with a ClearAndPrepareForLength() call.
MozReview-Commit-ID: 1A771gsHWan
Bug 1345294 introduced nsPrefBranch::{get,set}StringPref(), which allowed the
getting of utf8 strings from prefs, which previously required using
nsISupportsString with {get,set}ComplexValue. That bug also converted most
uses.
This patch finishes the job.
- It removes the nsISupportsString support.
- It converts existing code that relied on the nsISupportsString.
- It removes the lint that was set up to detect such uses of nsISupportsString.
libpref has callbacks. You can register a callback function against a
particular pref (or pref prefix), and it'll be called when any matching pref
changes.
This is implemented in two layers. The lower layer is the list of CallbackNode
objects, pointed to by gFirstCallback.
The upper layer involves gObserverTable, which is a hash table of ValueObserver
objects. It is used for callbacks registered via
Preferences::RegisterCallback() and Preferences::Add*VarCache(), but not for
observers registered with Preferences::Add{Weak,Strong}Observer(). If multiple
callbacks with identical prefnames, callback functions and MatchKinds occur,
they are commoned up into a single ValueObserver, which is then wrapped by a
single CallbackNode. (The callbacks within a single ValueObserver can have
different void* aClosure args; ValueObserver keeps those in an nsTArray.)
Note also that gObserverTable is very inelegant, with duplication of data
between the keys and the values due it being a
nsRefPtrHashtable<ValueObserverHashKey, ValueObserver> and ValueObserver being
a subclass of ValueObserverHashKey(!)
This extra layer might make sense if there were a lot of commoning up
happening, but there's not. Across all process kinds there is an average of
between 1.2 and 1.7 closures per ValueObserver. The vast majority have 1
closure, and there are a handful that have multiple; the highest number I've
seen is 71.
(Note that this two-layer design probably seemed more natural back when libpref
was spread across multiple files.)
This patch removes the ValueObserver layer. The only tricky part is that there
is a notion of a MatchKind -- ExactMatch or PrefixMatch -- which exists in the
ValueObserverLayer but not in the CallbackNode layer. So the patch moves the
MatchKind into the CallbackNode layer, which is straightforward.
On Linux64 this reduces memory usage by about 40 KiB in the parent process and
about 30 KiB in each content processes.
The performance implications are minor.
- The list of CallbackNodes is a bit longer, so when a pref is changed we must
check more nodes. But we avoid possibly doing prefname matching twice.
- Callback registration is much faster, just a simple list prepend, without any
hash table insertion required.
- Callback unregistration is probably not much different. There is a longer
list to traverse, but no hash table removal.
- Pref lookup is by far the hottest libpref operation, and it's unchanged.
For perf to be genuinely affected would require (a) a *lot* more
almost-duplicate callbacks that would have been commoned up, and (b) frequent
changing of the pref those callbacks are observing. This seems highly unlikely.
MozReview-Commit-ID: 6xo4xmytOf3
Bug 1345294 introduced nsPrefBranch::{get,set}StringPref(), which allowed the
getting of utf8 strings from prefs, which previously required using
nsISupportsString with {get,set}ComplexValue. That bug also converted most
uses.
This patch finishes the job.
- It removes the nsISupportsString support.
- It converts existing code that relied on the nsISupportsString.
- It removes the lint that was set up to detect such uses of nsISupportsString.
Pref callbacks registered via Preferences::Add*VarCache() are wrapped in a
ValueObserver -- which groups all callback requests that have the same prefname
and callback function -- and the ValueObserver is put into gObserverTable. This
is reasonable.
Observers registered via nsPrefBranch::Add{Weak,Strong}Observer() are wrapped
in a PrefCallback, and the PrefCallback is put into sRootBranch->mObservers.
This is also reasonable.
Pref callbacks registered via Preferences::RegisterCallback() are conceptually
similar to those registered via Preferences::Add*VarCache(). However, they are
implemented by using *both* of the above mechanisms: they are wrapped in a
ValueObserver which is put into gObserverTable, *and* that ValueObserver is then
wrapped in a PrefCallback which is put into sRootBranch->mObservers.
Using both mechanisms isn't necessary, so this patch removes the
PrefCallback/mObservers part. This makes Preferences::RegisterCallback() work
in much the same way as Preferences::Add*VarCache().
Specifically:
- Preferences::RegisterCallback() now calls PREF_RegisterCallback() instead of
Preferences::AddStrongObserver(). This makes it more similar to
RegisterPriorityCallback().
- Preferences::UnregisterCallback() now explicitly calls
PREF_UnregisterCallback() instead of Preferences::RemoveObserver (which
previously happened via ~ValueObserver() when the ValueObserver was removed
from gObserverTable and its refcount dropped to zerod).
MozReview-Commit-ID: 1tEQNeYrBUU
It's no longer needed, now that legacy extensions aren't supported.
Pieces removed include the following.
- The "load-extension-default" observer notification.
- The code for reading defaults/preferences/*.js from extensions.
- The unit test for this stuff.
- A crash reporter annotation relating to very long prefs set by add-ons.
- All references to "ExtPrefDL".
MozReview-Commit-ID: KMBoYn3uZ3x
The notable part of this change is Shutdown(). I've made it just null out
sPreferences, contrary to the old comment, which was strange for a couple of
reasons:
- ~Preferences() used to null out sPreference, which is backwards compared to
how this sort of thing normally works.
- In both the before and after cases, as far as I can tell,
Preferences::Shutdown() is called but ~Preferences() is never called;
something keeps the singleton Preferences instance alive until process
termination.
MozReview-Commit-ID: Ab0ui31rVcI
sRootBranch and sDefaultRootBranch have the same lifetime as sPreferences, so
this patch makes them non-static nsCOMPtr<> members of Preferences.
MozReview-Commit-ID: 1TLhh13ZpBI
This patch does the following.
- Reduces nesting and simplifies control flow by handling failure cases
earlier, and makes all three functions have identical structure.
- Avoids unnecessary temporary local variables, including |rv|.
- Removes low-value comments, including some misleading ones (e.g. "This
function will perform conversion...").
- PREF_GetCStringValue() previously did not check HasDefaultValue(), unlike the
other two. It now does.
- PREF_GetCStringValue() now calls SetIsVoid(true) at the start, so that
aValueOut will be Voided on all failure paths.
MozReview-Commit-ID: 2uCSv76Y8eu
