Extract of Python 2.7, Modules/audioop.c:

static int
fbound(double val, double minval, double maxval)
{
    if (val > maxval)
        val = maxval;
    else if (val < minval + 1)
        val = minval;
    return val;
}

Example of usage:

    double factor, fval, maxval, minval;
    int len, size, val = 0;

        fval = (double)val*factor;
        val = (int)floor(fbound(fval, minval, maxval));

It seems wrong to me to call floor() with an integer. Why fbound() doesn't return a double?

fbound() has been modified to explicitly cast its result to an int in the master branch:

static int
fbound(double val, double minval, double maxval)
{
    if (val > maxval)
        val = maxval;
    else if (val < minval + 1)
        val = minval;
    return (int)val;
}

But master still uses something like:

        val = floor(fbound(val, minval, maxval));


It seems like fbound() result is always passed into floor(). Maybe floor() should be moved into fbound()?

Attached PR changes fbound() to round correctly: call floor() into fbound().

--

I looked at the code because of the following compiler warning on Python 2.7:

[00:02:21] ..\Modules\audioop.c(38): warning C4244: 'return' : conversion from 'double' to 'int', possible loss of data [C:\projects\cpython\PCbuild\pythoncore.vcxproj]

Oh, fbound() cast to (int) has been done by... myself :-D

commit f2b9a340ef143099726fb642951410e4ad793c7f
Author: Victor Stinner <victor.stinner@gmail.com>
Date:   Tue May 7 23:49:15 2013 +0200

    audioop: explicit cast to fix a compiler warning

diff --git a/Modules/audioop.c b/Modules/audioop.c
index 7e40bbddc6..4f2b25f33a 100644
--- a/Modules/audioop.c
+++ b/Modules/audioop.c
@@ -37,7 +37,7 @@ fbound(double val, double minval, double maxval)
         val = maxval;
     else if (val < minval + 1)
         val = minval;
-    return val;
+    return (int)val;
 }


Stupid me!

New changeset 45e4efba7fa2abe61d25e4f8b5bf482e19ff1280 by Victor Stinner in branch 'master':
bpo-33781: audioop: enhance rounding double as int (GH-7447)
https://github.com/python/cpython/commit/45e4efba7fa2abe61d25e4f8b5bf482e19ff1280

It was my mistake. fbound() should be return double.

But since it is always used with floor(), it is better to move floor() inside it.

> But since it is always used with floor(), it is better to move floor() inside it.

Yeah, that's what I did ;-) I will backport the change to 3.7, 3.6 and 2.7. My final goal is just to fix a compiler warning in 2.7 :-D

New changeset 5c022f13ab6db8929e092ad035b3dc61701e3198 by Miss Islington (bot) in branch '3.7':
bpo-33781: audioop: enhance rounding double as int (GH-7447)
https://github.com/python/cpython/commit/5c022f13ab6db8929e092ad035b3dc61701e3198

Note: I chose to not add a NEWS entry, because even if my change might enhance the rounding, I don't think that it has any effect in practice :-)

New changeset b17d409bc0458e3981987c2358da661f228f5891 by Victor Stinner in branch '2.7':
bpo-33781: audioop: enhance rounding double as int (GH-7447) (GH-7452)
https://github.com/python/cpython/commit/b17d409bc0458e3981987c2358da661f228f5891

> New changeset b17d409bc0458e3981987c2358da661f228f5891 by Victor Stinner in branch '2.7':
> bpo-33781: audioop: enhance rounding double as int (GH-7447) (GH-7452)

I checked AppVeyor logs: this change fixed the audioop.c warning that I wanted to fix ;-)

New changeset 2bc1946fb02e140f5f5ac21a1afa2763615ad16b by Victor Stinner (Miss Islington (bot)) in branch '3.6':
bpo-33781: audioop: enhance rounding double as int (GH-7447) (GH-7451)
https://github.com/python/cpython/commit/2bc1946fb02e140f5f5ac21a1afa2763615ad16b

New changeset e5b79c546370521764457ea2ec809303e580f5ea by Victor Stinner in branch '2.7':
bpo-19418: audioop.c: Fix warnings on -0x80000000 (GH-7453)
https://github.com/python/cpython/commit/e5b79c546370521764457ea2ec809303e580f5ea

Ok, fbound() is now better in all branches. I also fixed all compiler warnings in 2.7. I close the issue.