Performance tests may not be capturing file open time in brackets-shell

[core-ai-bot]

Issue by peterflynn
Tuesday Sep 18, 2012 at 22:31 GMT
Originally opened as adobe/brackets#1687

---

Looking at our file-open performance data:
https://docs.google.com/spreadsheet/ccc?key=0Aras0diokeHxdEc5RGtOeVI0V0xGU3FPUXBuX3ZYTlE#gid=5

There was a dramatic drop in the recorded time to open large files when we cut over to brackets-shell. There's now almost no difference between different file sizes.

That seems a little suspicious given that most disk operations have seemed slightly slower in -shell. I'm wondering if the greater degree of asynchronicity could be exposing a bug in where we've placed the instrumentation. Seems worth investigating.

[core-ai-bot]

Comment by pthiess
Tuesday Sep 18, 2012 at 23:56 GMT

---

Hi@peterflynn I would eventually consider this a medium priority, we need to trust the performance measurements. What's your take?

[core-ai-bot]

Comment by peterflynn
Wednesday Sep 19, 2012 at 00:43 GMT

---

@pthiess: sure, medium seems fair. There's also a slight chance this is a symptom of a functional async bug too, so it definitely seems worth investigating.

However, IMHO we shouldn't really trust any of these performance measurements :-) Because they capture so little of the rendering/layout cost, which can be substantial. I think the camera tests suggested that these file-open measurements inaccurate by 10-30%, depending on file size.

[core-ai-bot]

Comment by pthiess
Wednesday Sep 19, 2012 at 01:16 GMT

---

Agreed, we probably have no handle on being accurate about the layout cost, maybe using a set of files to calculate a compound average could minimize the experimental variance.

[core-ai-bot]

Comment by pthiess
Wednesday Sep 19, 2012 at 01:20 GMT

---

@peterflynn@jasonsanjose Raised priority to medium.

[core-ai-bot]

Comment by pthiess
Wednesday Sep 26, 2012 at 22:18 GMT

---

Reviewed

[core-ai-bot]

Comment by jasonsanjose
Tuesday Oct 16, 2012 at 23:27 GMT

---

I'm not seeing any functional issues in the perf instrumentation as-is. I traced through and debugged the call stack to be sure that we do add the editor to the DOM synchronously before we stop our perf timer. I didn't notice the disk slowdown that@peterflynn mentioned when we moved to CEF3. Is there something else that I'm missing?

[core-ai-bot]

Comment by pthiess
Monday Oct 22, 2012 at 18:57 GMT

---

@jasonsanjose Maybe do a perceptional testing using an older build with brackets-app vs. the current.

[core-ai-bot]

Comment by lkcampbell
Sunday Jan 05, 2014 at 17:45 GMT

---

@jasonsanjose and@pthiess, this issue is really stale and doesn't seem like it is medium priority anymore. Do we ever reassign priorities based on the age of an issue? It seems like a really old medium priority issue should eventually turn into either a low priority issue or a high priority issue.

[core-ai-bot]

Comment by pthiess
Monday Jan 06, 2014 at 18:27 GMT

---

@lkcampbell Great question - I frequently ask the team to review their medium priority bugs.@jasonsanjose I wonder if we should just close this issue for now.