This is by no means perfect, but it gets the kernel tasks running on my
machine out of the profiles. We will no doubt need to add more in the
future, or find a way to record a flag for that in the capture format.
Unfortunately we don't have a good way to do this in GtkScrolledWindow
where you have an initial gravity pulling you to the bottom of a viewport.
We might be able to fake it with a focusable widget, but that sounds like
more work than just animating to the bottom of the viewport.
Related #93
If we get a request for a process that we have not captured any information
about then give it the "Unknown Process" symbol. That way we do not crash
and we also maintain our invariant of not mutating the hash table.
If we get an empty string, just normalize that to NULL so that we can be
more likely to match equality checks via hash comparison.
Additionally, break hashes out into two so that we can improve the
situation where some symbols do not have paths but still match. This
can happen with bundled symbols.
We only mutate this during loading of the document so that we can be
confident in multi-threaded workers after loading. This just asserts
that invariant holds true.
This is useful for when we are viewing things like flamegraphs which can
take advantage of options in the callgraph.
Also add some labels for the header bar buttons.
This makes the graphs interactive finally where you can select a node and
see the children of it directly. Click down the stack to return to a
larger view.
This helps with the situation where we are going to render the parents
of the selected root first, and want them to take the whole width. Next
step is to support invalidation of the rendering.
You can click on flamegraph items (and soon) we will update the graph to
show you that node. Set the hand cursor when doing that so the user knows
that is an option.
Now that we do some filtering on these to not show too many items per row,
this can be enabled. Previously it could *really* slow things down due to
all the app<->compositor traffic it causes.
We don't want to cost-account the same summary multiple times while walking
up to the root. Otherwise, you can get items which come out to a percentage
of > 100% which is not exactly what you're expecting to see as a normalized
value.
This represented a large stall when loading the window, and also results
in doing a bunch of work twice as we set the model (then again the time
range).
So instead, just do it incrementally and let the functions list backfill in
a bit.
This is something that original flamegraphs do to aid in seeing adjacent
towers. We want that too, but we need it to be stable across redraws. Use
the hash of the symbol rather than g_random_double_range() for that.
This makes things look a bit more like flamegraphs.pl in the sense that we
have some labels and separation between rows. Also, use a ScrolledWindow so
that we can have much taller graphs to accommodate deep stack traces.
We might want to jump to the bottom at some point, but this gets things in
place for now. Icicle graphs are another option (invert).
This shouldn't affect categorizing because that only uses the value if
is_toplevel. But with this added, we can use the count for weights in
other tooling w/o needing augmentation.
