RejectedSoftware Forums

On Tue, 11 Feb 2014 11:16:47 +0100, Sönke Ludwig wrote:

Am 10.02.2014 22:58, schrieb Stephan Dilly:

i already posted this in the d NG:

I gave beta 3 on my win32 dev machine a try today and noticed
that building vibe.d (a dependancy in almost all my projects)
takes noticeably longer to built. observing the process shows
that beta3 consumes almost 1.5GB of RAM while dmd 2.064 'just'
uses 1GB.

I am not so familiar with the compiler code and what could be causing this ? can anyone else confirm this ?

I did notice some out-of-memory errors for projects that compiled fine
before, so that's probably the same cause. But I didn't do anything
because I was (and still am) busy with other compiler regressions.

If the D NG post doesn't result in a solution, I think a regression bug
report may be in order. DMD's memory usage has always been quite high
and if anything, it must do down, not up. It's already been a constant
fight for me to somehow get the bigger projects with Diet templates to
build over the last 1.5 years and a sudden degradation by 150% is a lot.

Problem is that I get this degradation even with the simplest hello world vibe.d application without using ANY diet templates. this must be addressed befor the next dmd release, letting this sneak in and wait another 4months for the next release would be a disaster

On Tue, 11 Feb 2014 15:03:26 GMT, Etienne wrote:

On Tue, 11 Feb 2014 11:03:17 GMT, Stephan Dilly wrote:

Problem is that I get this degradation even with the simplest hello world vibe.d application without using ANY diet templates. this must be addressed befor the next dmd release, letting this sneak in and wait another 4months for the next release would be a disaster

I second that. At least DMD should know about this flaw and throw cleanly when out of memory, because currently the O/S takes care of that and just crashes everything related to compilation when in debian. It's quite a disappointing bug. On the plus side, I've never seen it go beyond 2gb of memory use even with dozens of diet templates

Yeah here it ends with ~1.5GB which I guess is the threshold because it is 32bt dmd :D

On Tue, 11 Feb 2014 15:12:56 GMT, Stefan Koch wrote:

On Tue, 11 Feb 2014 15:03:26 GMT, Etienne wrote:

On the plus side, I've never seen it go beyond 2gb of memory use even with dozens of diet templates.
That is because the ctfe overhead is so big everything else is noise ...

I prefer a big ctfe overhead than a big source code :P

On Mon, 10 Feb 2014 21:58:59 GMT, Stephan Dilly wrote:

i already posted this in the d NG:

I gave beta 3 on my win32 dev machine a try today and noticed 
that building vibe.d (a dependancy in almost all my projects) 
takes noticeably longer to built. observing the process shows 
that beta3 consumes almost 1.5GB of RAM while dmd 2.064 'just' 
uses 1GB.

I am not so familiar with the compiler code and what could be causing this ? can anyone else confirm this ?

Confirmed, it totally debilitates my 1GB RAM computer with 1GB of swap (taking, as you said, about 1.5GB) whereas with 2.064 it wasn't as bad. Unfortunately I need to use 2.065b3 to avoid a bug. Hopefully the reason behind this can be found (and fixed).