Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
RE: [cdt-dev] CDT performance

I have created bugzilla ID 115993 to track proposal with precompiled
headers.
https://bugs.eclipse.org/bugs/show_bug.cgi?id=115993 
Comments are welcome.
-----------------------------------
With best regards, Oleg Krasilnikov
Software designer, Eclipse team. 
Intel corp.
+7 8312 162 444 ext. 2587 
(Russia, Nizhny Novgorod)

 
-----Original Message-----
From: cdt-dev-bounces@xxxxxxxxxxx [mailto:cdt-dev-bounces@xxxxxxxxxxx]
On Behalf Of Krasilnikov, Oleg
Sent: Monday, November 07, 2005 5:41 PM
To: CDT General developers list.
Cc: Voronin, Mikhail; Sennikovsky, Mikhail
Subject: [cdt-dev] CDT performance

 
Hi all.

Please find in attach a proposal for precompiled headers.

This feature was intended to speed up indexing process.

But it conflicts with proposed PDOM implementation, so
it seems to be not reasonable to implement precompiled
headers feature on a basis of current code. But, may be,
it can be however implemented in future, side by side 
with PDOM. Detais are in attached document.

By the way. Final patch for CEditor speed-up is put to:
https://bugs.eclipse.org/bugs/show_bug.cgi?id=113518 
-----------------------------------
With best regards, Oleg Krasilnikov
Software designer, Eclipse team. 
Intel corp.
+7 8312 162 444 ext. 2587 
(Russia, Nizhny Novgorod)

 
-----Original Message-----
From: cdt-dev-bounces@xxxxxxxxxxx [mailto:cdt-dev-bounces@xxxxxxxxxxx]
On Behalf Of Krasilnikov, Oleg
Sent: Monday, October 24, 2005 2:17 PM
To: CDT General developers list.
Cc: Voronin, Mikhail; Sennikovsky, Mikhail
Subject: RE: [cdt-dev] CDT performance: 

Hi all.

I propose a small modification for CDT performance improvement in Cmodel
implementation.

Currently, C/C++ source file is parsed again and again after each
modification in CEditor.

So, I propose:

After "documentChanged" event fires, check whether modification 
is located in global space or inside of any function body. 
In first case, run "CReconcilingStrategy.reconcile()", as usual. 
Else, do nothing, since these changes are not meanful for outline.

Test results on 200K file:
Before: 230-420 ms for parsing after each typing.
After: less than 1 ms normally. In rare cases, max 10 ms for additional
checks + usual parsing time.

Bug 113518 has been added to the database, it contains detail
description.
https://bugs.eclipse.org/bugs/show_bug.cgi?id=113518 

-----------------------------------
With best regards, Oleg Krasilnikov
Software designer, Eclipse team. 
Intel corp.
+7 8312 162 444 ext. 2587 
(Russia, Nizhny Novgorod)


 
-----Original Message-----
From: cdt-dev-bounces@xxxxxxxxxxx [mailto:cdt-dev-bounces@xxxxxxxxxxx]
On Behalf Of Krasilnikov, Oleg
Sent: Thursday, October 20, 2005 8:00 PM
To: CDT General developers list.
Cc: Voronin, Mikhail
Subject: [cdt-dev] CDT performance 

Hi all.

Can anybody say whether the following things (related to CDT
parformance) were discussed ?

1. CModel optimization: currently, whole C file is parsed after each
change in editor.
   May be it's reasonable to re-parse changed fragment only. 
   Problem: even if 1 char changed, whole code structure can be affected
in some cases.
   Mean to resolve: if we are "not sure", we can re-parse whole file,
like now. In worst case,
   we'll have the same situation as now (additional checks seem to be
low-weight).

2. Store source files (expecially common includes) in some "half-parsed"
format.
   In general, 
     - create "mirror" directory (possibly on fastest hard disc, unlike
slow network CVS devices)
     - for each original source file, create corresponding mirror file,
which should contain:
         - macros definitions
         - includes definitions
         - token objects
         - unresolved macros  
         - specific records for #ifdef / #if / ... statements
   When scanner references some source file, we can check whether mirror
file is newer
   than original one. If so, we can use mirror file - so, no need to do
scanning again: we
   are skipping most of scanner activity (char-by char text analyse) 
   Disadvantage: need to write mirror file after each source change.
   May be, it's reasonable to enable mirroring for specific directories
only (system includes,
   which are not intended for change but referenced multiple times from
different sources).

I'm investigating these things now, but may be somebody have already
worked with it ?
-----------------------------------
With best regards, Oleg Krasilnikov
Software designer, Eclipse team. 
Intel corp.
+7 8312 162 444 ext. 2587 
(Russia, Nizhny Novgorod)
_______________________________________________
cdt-dev mailing list
cdt-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-dev
_______________________________________________
cdt-dev mailing list
cdt-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cdt-dev


Back to the top