From: Yohanes Santoso <ysantoso-rubycore@...>
Date: 2006-02-01T06:07:58+09:00
Subject: Re: PATCH: append option to sysread -- retracted

Tanaka Akira <akr@m17n.org> writes:

> In article <87bqxt6xy8.fsf@jenny-gnome.dyndns.org>,
>   Yohanes Santoso <ysantoso-rubycore@jenny-gnome.dyndns.org> writes:
>
>> Otherwise, the original code:
>>
>>    n = read(fileno(fptr->f), RSTRING(str)->ptr, ilen);
>>
>> could also corrupt memory because another thread may have resize str
>> to less than ilen.
>
> There is check code.
>
>     if (RSTRING(str)->len != ilen) {
>         rb_raise(rb_eRuntimeError, "buffer string modified");
>     }
> -- 
> Tanaka Akira

That check code has been updated in the patch too:

    if (RSTRING(str)->len != initial_len + ilen) {
        rb_raise(rb_eRuntimeError, "buffer string modified");
    }

1. Is that sufficient? 

I can't produce a test case where I could modify (including
resizing) the string without some exception being thrown, thus
corrupting memory as asked in your original question.


2. In any case, I am retracting this patch because I and one other
person are not getting consistent performance improvement with this
patch.

My initial finding seems to point to the linear-growth of the string
and also at the mercy of the numerous realloc() being performed.

I'll play with exponential-growing string.


YS.