Heap dump overhaul - Githubissues

petrcermak commented 8 years ago

petrcermak commented 8 years ago

To make the heap dump UI intuitive, I would like to display small icons in the tabs (ƒ and Ⓣ):

screenshot from 2016-08-31 16 13 59

To do this, I uploaded a patch adding the required functionality to (https://codereview.chromium.org/2292293003):

screenshot from 2016-08-31 16 11 58

petrcermak commented 8 years ago

I built a prototype of the new UI last week: https://codereview.chromium.org/2300013002

petrcermak commented 8 years ago

Here's a list of all the feature requests and prototype comments I received so far:

[ ] The column sort order should be kept when moving back and forth between levels. [@mariakhomenko]
[ ] Remember what has been exapanded and selected and attempt to restore the view when a different memory dump is selected. [mmenke@]
[ ] Sort by stack frame by default. The current default order is random, which seems not terribly useful. [mmenke@]

[x] By default, size column on the right isn't visible. I thought it wasn't there until I resized left side and there's no horizontal scrollbar indicator within view either (it's only visible if you scroll all the way down). [@mariakhomenko]
[ ] It's a bit odd that the whole analysis view scrolls instead of the bottom right pane. I think that panes should scroll individually, keeping the overview on top. [@mariakhomenko]
[ ] The lack of bottom border looks pretty odd on short stacks—kinda looks like page stopped rendering. I think adding a bottom border would look better. [@mariakhomenko]

[ ] Add a button to expand 'most heavy trace' would be nice. I'm usually interested in stack traces that contribute the most. [dskiba@]
[ ] Subtract tracing_overhead from the top-level value in the heap dump. [dskiba@]
[ ] Stack fusion (displaying "a → (b-c-d as one node) → e" instead of "a → b → c → d → e" if b and c each have only a single child) is one of the most requested features of any sampling tools we build. [@natduca]

[ ] Add support for showing hierarchy of categories, without any stack traces or types. [dskiba@]
[ ] Compute real categories by just walking the backtrace, stopping at first non-base non-STL frame and infer category from its path. This will give more precise categories. Another variation is to use namespace as a category, but follow the same pattern (pick first non-"base" thing). [dskiba@]

petrcermak commented 8 years ago

Delta sparklines (dependency) are now supported: #2788

petrcermak commented 8 years ago

Preventing changing table sort order and step-into event (dependencies) are now supported: #2751

petrcermak commented 8 years ago

Version 1 of the new UI is in review (https://codereview.chromium.org/2347813002) :tada:

screenshot from 2016-09-16 10 14 55

catapult-project / catapult