Re: Memory again

[Carsten Mattner]

On Wed, Dec 7, 2011 at 3:01 PM, Stefan Monnier <address@hidden> wrote:
>>> > You are, in fact, saying that fontifying a 60-MByte compilation buffer
>>> > creates text properties and other supporting data structures that can
>>> > accrue to a total of 2G bytes.  I find this extremely hard to believe.
>>> > A simple calculation will show that a file that has 1 million lines
>>> > with 5 different faces (warning, line number, column number,
>>> > underline, default) per line needs only a few million intervals -- how
>>> > can this require 2G bytes of memory?
>
> Because the property values are themselves non-trivial.
>
>>> Start fresh Emacs instance, make sure you're in *scratch*. Type M-x
>>> compilation-minor-mode, then eval (staying in *scratch*, of course):
>>>
>>> (let ((x 0)) (while (< x 1000000) (insert "test:1:oops\n") (setq x (1+ x))))
>>>
>>> Wait for completion, then see RSS. Next, kill *scratch*, M-x garbage-collect
>>> and see RSS again.
>
>> How is this different from compiling a C file with 1000000 #warning
>> directives?  What am I missing?
>
> To test your buffer-reallocation theory, you could instead do
>
> (insert (apply #'concat (make-list 1000000 "test:1:oops\n")))

Again, totally unscientific test without real instruments.

On a fresh trunk --with-ns build:

paste "(insert (apply #'concat (make-list 1000000 "test:1:oops\n")))"
to *scratch* and M-x eval-buffer: from 21.8MB to 55.8MB

as this was *scratch*, open ~/.bashrc in another buffer and kill both
and it's down to 44.4MB

M-x garbage-collect 32.9MB

M-x garbage-collect, M-x garbage-collect: 29.2MB