digitalmars.D - Re: Integer conversions too pedantic in 64-bit
- Kagamin <spam here.lot> Feb 17 2011
- dsimcha <dsimcha yahoo.com> Feb 17 2011
- Kagamin <spam here.lot> Feb 17 2011
- bearophile <bearophileHUGS lycos.com> Feb 17 2011
- "Steven Schveighoffer" <schveiguy yahoo.com> Feb 17 2011
- "Steven Schveighoffer" <schveiguy yahoo.com> Feb 17 2011
dsimcha Wrote:Now that DMD has a 64-bit beta available, I'm working on getting a whole bunch of code to compile in 64 mode. Frankly, the compiler is way too freakin' pedantic when it comes to implicit conversions (or lack thereof) of array.length. 99.999% of the time it's safe to assume an array is not going to be over 4 billion elements long. I'd rather have a bug the 0.001% of the time than deal with the pedantic errors the rest of the time, because I think it would be less total time and effort invested. To force me to either put casts in my code everywhere or change my entire codebase to use wider integers (with ripple effects just about everywhere) strikes me as purity winning out over practicality.
int ilength(void[] a) property { return cast(int)a.length; } --- int mylen=bb.ilength; ---
Feb 17 2011
Funny, as simple as it is, this is a great idea for std.array because it shortens the verbose cast(int) a.length to one extra character. You could even put an assert in it to check in debug mode only that the conversion is safe. On 2/17/2011 7:18 AM, Kagamin wrote:dsimcha Wrote:Now that DMD has a 64-bit beta available, I'm working on getting a whole bunch of code to compile in 64 mode. Frankly, the compiler is way too freakin' pedantic when it comes to implicit conversions (or lack thereof) of array.length. 99.999% of the time it's safe to assume an array is not going to be over 4 billion elements long. I'd rather have a bug the 0.001% of the time than deal with the pedantic errors the rest of the time, because I think it would be less total time and effort invested. To force me to either put casts in my code everywhere or change my entire codebase to use wider integers (with ripple effects just about everywhere) strikes me as purity winning out over practicality.
int ilength(void[] a) property { return cast(int)a.length; } --- int mylen=bb.ilength; ---
Feb 17 2011
dsimcha Wrote:Funny, as simple as it is, this is a great idea for std.array because it shortens the verbose cast(int) a.length to one extra character. You could even put an assert in it to check in debug mode only that the conversion is safe.
int ilength(void[] a) property { return cast(int)a.length; }
I'm not sure the code is correct. I have a vague impression that void[] is like byte[], at least, it's used as such, and conversion from int[] to byte[] multiplies the length by 4.
Feb 17 2011
Steven Schveighoffer:Yes, David has proposed a corrected version on the Phobos mailing list: http://lists.puremagic.com/pipermail/phobos/2011-February/004493.html
I suggest it to return a signed value, like an int. But a signed long is OK too. I suggest a name as "len" (or "slen") because I often write "length" wrongly. Does it support code like: auto l = arr.len; arr.len = 10; arr.len++; A big problem: it's limited to arrays, so aa.len or rbtree.len, set.len, etc, don't work. So I'd like something more standard... So I am not sure this is a good idea. Bye, bearophile
Feb 17 2011
On Thu, 17 Feb 2011 09:45:14 -0500, Kagamin <spam here.lot> wrote:dsimcha Wrote:Funny, as simple as it is, this is a great idea for std.array because it shortens the verbose cast(int) a.length to one extra character. You could even put an assert in it to check in debug mode only that the conversion is safe.
int ilength(void[] a) property { return cast(int)a.length; }
I'm not sure the code is correct. I have a vague impression that void[] is like byte[], at least, it's used as such, and conversion from int[] to byte[] multiplies the length by 4.
Yes, David has proposed a corrected version on the Phobos mailing list: http://lists.puremagic.com/pipermail/phobos/2011-February/004493.html -Steve
Feb 17 2011
On Thu, 17 Feb 2011 13:08:08 -0500, bearophile <bearophileHUGS lycos.com> wrote:Steven Schveighoffer:Yes, David has proposed a corrected version on the Phobos mailing list: http://lists.puremagic.com/pipermail/phobos/2011-February/004493.html
I suggest it to return a signed value, like an int. But a signed long is OK too. I suggest a name as "len" (or "slen") because I often write "length" wrongly.
This isn't replacing length, it is in addition to length (which will continue to return size_t).Does it support code like: auto l = arr.len; arr.len = 10; arr.len++;
arr.length = 10 already works. It's int l = arr.length that doesn't. if arr.length++ doesn't work already, it should be made to work (separate bug).A big problem: it's limited to arrays, so aa.len or rbtree.len, set.len, etc, don't work. So I'd like something more standard... So I am not sure this is a good idea.
The point is to avoid code like cast(int)arr.length everywhere you can safely assume arr.length can fit in a (u)int. This case is extremely common for arrays, you seldom have an array of more than 2 or 4 billion elements. For other types, the case might not be as common, plus you can add properties to other types, something you cannot do to arrays. As far as I'm concerned, this isn't going to affect me at all, I like to use size_t. But I don't see the harm in adding it. -Steve
Feb 17 2011