Re: [Pharo-dev] Faster ThreadSafeTranscript Re: [pharo-project/pharo-core] 06d05b: 40165

[Ben Coman]

Eliot Miranda wrote: On Tue, Aug 12, 2014 at 5:27 AM, Ben Coman <b...@openinworld.com> wrote: GitHub wrote: Branch: refs/heads/4.0 Home: https://github.com/pharo-project/pharo-core Commit: 06d05bd822deee4a79736d9f99d4a666ca1637eb https://github.com/pharo-project/pharo-core/commit/06d05bd822deee4a79736d9f99d4a666ca1637eb Author: Jenkins Build Server <bo...@pharo-project.org> Date: 2014-08-11 (Mon, 11 Aug 2014) 13806 Remove ThreadSafeTranscriptPluggableTextMorph https://pharo.fogbugz.com/f/cases/13806 For anyone concerned about the performance of writing to Transcript from higher priority threads, just reporting that altering ThreadSafeTranscript to be safe for Morphic without ThreadSafeTranscriptPluggableTextMorph had a side effect of enhancing performance by 25x.   With two runs of the following script...     Smalltalk garbageCollect.     Transcript open. "close after each run"     [    Transcript crShow: (              [             | string |             string := '-'.                1 to: 2000 do:             [        :n |                      string := string , '-', n printString.                     Transcript show: string.             ].             (Delay forMilliseconds: 10) wait.         ]) timeToRun.     ] forkAt: 41. Build 40162 reports timeToRun of 0:00:00:02.483 & 0:00:00:02.451 Build 40165 reports timeToRun of 0:00:00:00.037 & 0:00:00:00.099 Now I had meant to ask... I notice that FLFuelCommandLineHandler installs ThreadSafeTranscript, so I wonder it is affected by this change. Can some Fuel experts comment? Also I am looking for some advice for a minor downside I just noticed.  The whole script above can complete between steps, so the entire output ends up in the PluggableTextMorph its size without being culled, which causes making a selection become really slow.  Normally the excess text shown by Transcript is culled in half [1] by PluggableTextMorph>>appendEntry each time #changed: is called. PluggabletextMorph>>appendEntry     "Append the text in the model's writeStream to the editable text. "     textMorph asText size > model characterLimit ifTrue:   "<---[0]"         ["Knock off first half of text"         self selectInvisiblyFrom: 1 to: textMorph asText size // 2.   "<---[1]"         self replaceSelectionWith: Text new].     self selectInvisiblyFrom: textMorph asText size + 1 to: textMorph asText size.     self replaceSelectionWith: model contents asText.  "<----[2]"     self selectInvisiblyFrom: textMorph asText size + 1 to: textMorph asText size That works fine when #appendEntry is being called with lots of small changes, but for a single large change the entire change ends up in PluggableTextMorph via [2]. In this case     model characterLimit  "--> 20,000"     [0]     model contents size "--> 5,671,343"    [2] where model == Transcript. So what is the behaviour you'd like to when too much is sent to Transcript? a. Show all content however briefly. b. Only the last 20,000 characters are put into the PluggableTextMorph, and the earlier data thrown away. I see a few ways to deal with this: 1. Limit the stream inside Transcript to a maximum 20,000 characters by basing it on some circular buffer. 2. Have "Transcript contents" return only the last 20,000 characters of its stream. 3. Limit to text sent to #replaceSelectionWith: [2]  to 20,000 characters. Thoughts anyone? IMO 20,000 characters is way too little.  I routinely set my limit to 200,000.   So b) but with a higher limit.  a) is pointlessly slow. Thanks Eliot. Its good to know what is expected from Transcript. In your implementation list, surely 2. is the only thing that makes sense? I agree, (2.) would be efficient.  It just would have been precluded if (a) was required. If the limit can remain soft then anyone that wants to see more simply raises their limit.  Do you mean you'd like Transcript characterLimit to be a Preference ? 1. & 3. are implementation details left up to you.  I can do (1.) with Steph's new circular list.  Doing (2.) means (3.) wont be required for this, but its probably beneficial so I'll log it as separate Issue. As long as the transcript is reasonably fast and has the same protocol as WriteStream (it needs to support printOn: and storeOn: et al) implement it as you will.  BTW, the Squeak transcript is appallingly slow. -- best, Eliot