I don't have Perl on Windows, so I can't check whether all binary characters are read correctly, but I can confirm that on Mac OS X 10.4.9 Perl 5.8.6
<:raw:perlio does not lose buffering. I copied the PerlIO benchmark you referenced and added
raw:perlio. The results are below.
#### Testing for 1 KB Read
Benchmark: running mmap, nolayer, perlio, raw, raw:perlio for at least
+ 3 CPU seconds...
mmap: 29 wallclock secs ( 1.21 usr + 2.00 sys = 3.21 CPU) @ 89
+4.70/s (n=2872)
nolayer: 56 wallclock secs ( 1.52 usr + 1.67 sys = 3.19 CPU) @ 16
+57.05/s (n=5286)
perlio: 41 wallclock secs ( 1.58 usr + 1.53 sys = 3.11 CPU) @ 14
+14.79/s (n=4400)
raw: 7 wallclock secs ( 0.66 usr + 2.57 sys = 3.23 CPU) @ 12
+8.79/s (n=416)
raw:perlio: 51 wallclock secs ( 1.52 usr + 1.60 sys = 3.12 CPU) @ 15
+43.59/s (n=4816)
Rate raw mmap perlio raw:perlio nolay
+er
raw 129/s -- -86% -91% -92% -9
+2%
mmap 895/s 595% -- -37% -42% -4
+6%
perlio 1415/s 999% 58% -- -8% -1
+5%
raw:perlio 1544/s 1099% 73% 9% -- -
+7%
nolayer 1657/s 1187% 85% 17% 7%
+--
#### Testing for 4 KB Read
Benchmark: running mmap, nolayer, perlio, raw, raw:perlio for at least
+ 3 CPU seconds...
mmap: 28 wallclock secs ( 1.48 usr + 1.77 sys = 3.25 CPU) @ 84
+2.77/s (n=2739)
nolayer: 44 wallclock secs ( 1.66 usr + 1.39 sys = 3.05 CPU) @ 13
+60.66/s (n=4150)
perlio: 37 wallclock secs ( 1.81 usr + 1.37 sys = 3.18 CPU) @ 11
+85.53/s (n=3770)
raw: 5 wallclock secs ( 0.61 usr + 2.57 sys = 3.18 CPU) @ 33
+.96/s (n=108)
raw:perlio: 41 wallclock secs ( 1.78 usr + 1.42 sys = 3.20 CPU) @ 12
+79.38/s (n=4094)
Rate raw mmap perlio raw:perlio nolay
+er
raw 34.0/s -- -96% -97% -97% -9
+8%
mmap 843/s 2381% -- -29% -34% -3
+8%
perlio 1186/s 3391% 41% -- -7% -1
+3%
raw:perlio 1279/s 3667% 52% 8% -- -
+6%
nolayer 1361/s 3906% 61% 15% 6%
+--
#### Testing for 16 KB Read
Benchmark: running mmap, nolayer, perlio, raw, raw:perlio for at least
+ 3 CPU seconds...
mmap: 19 wallclock secs ( 1.71 usr + 1.30 sys = 3.01 CPU) @ 52
+6.25/s (n=1584)
nolayer: 26 wallclock secs ( 2.11 usr + 1.07 sys = 3.18 CPU) @ 77
+6.73/s (n=2470)
perlio: 25 wallclock secs ( 2.17 usr + 1.04 sys = 3.21 CPU) @ 69
+3.15/s (n=2225)
raw: 4 wallclock secs ( 0.57 usr + 2.46 sys = 3.03 CPU) @ 8
+.58/s (n=26)
raw:perlio: 27 wallclock secs ( 2.14 usr + 1.04 sys = 3.18 CPU) @ 73
+3.96/s (n=2334)
Rate raw mmap perlio raw:perlio nolay
+er
raw 8.58/s -- -98% -99% -99% -9
+9%
mmap 526/s 6033% -- -24% -28% -3
+2%
perlio 693/s 7978% 32% -- -6% -1
+1%
raw:perlio 734/s 8453% 39% 6% -- -
+6%
nolayer 777/s 8952% 48% 12% 6%
+--
#### Testing for 64 KB Read
Benchmark: running mmap, nolayer, perlio, raw, raw:perlio for at least
+ 3 CPU seconds...
mmap: 12 wallclock secs ( 2.46 usr + 0.78 sys = 3.24 CPU) @ 24
+1.36/s (n=782)
nolayer: 14 wallclock secs ( 2.44 usr + 0.69 sys = 3.13 CPU) @ 29
+5.85/s (n=926)
perlio: 12 wallclock secs ( 2.41 usr + 0.67 sys = 3.08 CPU) @ 26
+8.18/s (n=826)
raw: 5 wallclock secs ( 0.60 usr + 2.64 sys = 3.24 CPU) @ 2
+.16/s (n=7)
raw:perlio: 13 wallclock secs ( 2.39 usr + 0.65 sys = 3.04 CPU) @ 28
+1.25/s (n=855)
Rate raw mmap perlio raw:perlio nolay
+er
raw 2.16/s -- -99% -99% -99% -9
+9%
mmap 241/s 11071% -- -10% -14% -1
+8%
perlio 268/s 12313% 11% -- -5% -
+9%
raw:perlio 281/s 12918% 17% 5% -- -
+5%
nolayer 296/s 13593% 23% 10% 5%
+--
#### Testing for 256 KB Read
Benchmark: running mmap, nolayer, perlio, raw, raw:perlio for at least
+ 3 CPU seconds...
mmap: 6 wallclock secs ( 2.61 usr + 0.39 sys = 3.00 CPU) @ 76
+.67/s (n=230)
nolayer: 7 wallclock secs ( 2.61 usr + 0.52 sys = 3.13 CPU) @ 85
+.30/s (n=267)
perlio: 7 wallclock secs ( 2.71 usr + 0.52 sys = 3.23 CPU) @ 76
+.16/s (n=246)
raw: 4 wallclock secs ( 0.71 usr + 3.05 sys = 3.76 CPU) @ 0
+.53/s (n=2)
(warning: too few iterations for a reliable count)
raw:perlio: 7 wallclock secs ( 2.72 usr + 0.51 sys = 3.23 CPU) @ 78
+.64/s (n=254)
Rate raw perlio mmap raw:perlio nola
+yer
raw 0.532/s -- -99% -99% -99% -
+99%
perlio 76.2/s 14218% -- -1% -3% -
+11%
mmap 76.7/s 14313% 1% -- -3% -
+10%
raw:perlio 78.6/s 14684% 3% 3% --
+-8%
nolayer 85.3/s 15937% 12% 11% 8%
+ --
