Spring Cleaning, and Load speedup #4515
Conversation
|
This is great. I saw the set mode slowdown when profiling a many-cell load. That and CM instantiation are the main issues loading large notebooks. |
|
JS tests look like real failure. |
| @@ -789,8 +789,8 @@ var IPython = (function (IPython) { | |||
| * | |||
| * @return cell {cell|null} created cell or null | |||
| **/ | |||
| Notebook.prototype.insert_cell_at_index = function(type, index){ | |||
|
|
|||
| Notebook.prototype.insert_cell_at_index = function(type, index, opts){ | |||
There was a problem hiding this comment.
In my dual mode editor branch #3605 I completely rework the cell selection logic and fix this as well - although in a different manner. Can we leave it out of this PR as it will make that PR more difficult to rebase.
I'm not sure it's instantiation, but the combinaison (instantiation+set_text) that triggers 2 redraw. I'm wondering if we can directly create a sell from JSON, instead of creating and then fromJSON. |
|
Removing the duplicate set_mode is definitely a great improvement, and buys us some time. I think with Brian's modal change, we can consider lazy instantiation, where we don't instantiate CM until the cell is focused for the first time. |
And how would you show code for code cell ? , use a 1pass highlight ? it will not be faster. @ellisonbg I removed the part you asked me to remove, but it is one of the major cause of |
ghost
assigned tab was not cancelling tooltip bringing to cases where you could have tooltip andcompleter open. Do not store '*' when serializing cells.
|
Rebased on master for conflict. |
1) Do not setOption('mode',new_mode) on CM if new and old mode are the
same. It triggert **a lot** of calculation of bounding box in the
end.
2) Do **not** select cell when loading the notebook it triggers
**a lot** of CM even that check visible things and so on and so
forth. So add a option to add_cell_at_index not to select it
3) jQuery $.attr has some magics, but has a slight overhead on
real native ELEM.setAttribute DOM method. Seem slight improvement
when loads of PNGs on one page
optimisation are not obvious, but order and time of attribute creation in javascript have impact for VMs apparently (Google IO talk on V8)
do not add the option not to select each cell in a row when buildogn the notebook at load time.
| @@ -576,12 +576,13 @@ var IPython = (function (IPython) { | |||
|
|
|||
| OutputArea.prototype.append_png = function (png, md, element) { | |||
| var toinsert = this.create_output_subarea(md, "output_png"); | |||
| var img = $("<img/>").attr('src','data:image/png;base64,'+png); | |||
| var img = $("<img/>") | |||
|
Looks good. I'll try to play with it and merge today. |
|
This is working smoothly. Definitely better, but still plenty of room for improvement. Merging. |
Spring Cleaning, and Load speedup
0) prompt '*' strore fix + tab remove tooltip
tab was not cancelling tooltip bringing to cases where you could have
tooltip andcompleter open.
Do not store '*' when serializing cells.
get rid of most slowdown at notebook loading.
1) Do not setOption('mode',new_mode) on CM if new and old mode are the
same. It triggert a lot of calculation of bounding box in the
end.
2) Do not select cell when loading the notebook it triggers
a lot of CM even that check visible things and so on and so
forth. So add a option to add_cell_at_index not to select it
3) jQuery $.attr has some magics, but has a slight overhead on
real native ELEM.setAttribute DOM method. Seem slight improvement
when loads of PNGs on one page
Spring Cleaning, and Load speedup
0) prompt '*' strore fix + tab remove tooltip
tab was not cancelling tooltip bringing to cases where you could have
tooltip andcompleter open.
Do not store '*' when serializing cells.
get rid of most slowdown at notebook loading.
1) Do not setOption('mode',new_mode) on CM if new and old mode are the
same. It triggert a lot of calculation of bounding box in the
end.
2) Do not select cell when loading the notebook it triggers
a lot of CM even that check visible things and so on and so
forth. So add a option to add_cell_at_index not to select it
3) jQuery $.attr has some magics, but has a slight overhead on
real native ELEM.setAttribute DOM method. Seem slight improvement
when loads of PNGs on one page
tab was not cancelling tooltip bringing to cases where you could have
tooltip andcompleter open.
Do not store '*' when serializing cells.
get rid of most slowdown at notebook loading.
Do not setOption('mode',new_mode) on CM if new and old mode are the
same. It triggert a lot of calculation of bounding box in the
end.
Do not select cell when loading the notebook it triggers
a lot of CM even that check visible things and so on and so
forth. So add a option to add_cell_at_index not to select it
jQuery $.attr has some magics, but has a slight overhead on
real native ELEM.setAttribute DOM method. Seem slight improvement
when loads of PNGs on one page