utf8: handle systems that don't write BOM for UTF-16
diff mbox series

Message ID 20190209200802.277139-1-sandals@crustytoothpaste.net
State New
Headers show
Series
  • utf8: handle systems that don't write BOM for UTF-16
Related show

Commit Message

brian m. carlson Feb. 9, 2019, 8:08 p.m. UTC
When serializing UTF-16 (and UTF-32), there are three possible ways to
write the stream. One can write the data with a BOM in either big-endian
or little-endian format, or one can write the data without a BOM in
big-endian format.

Most systems' iconv implementations choose to write it with a BOM in
some endianness, since this is the most foolproof, and it is resistant
to misinterpretation on Windows, where UTF-16 and the little-endian
serialization are very common. For compatibility with Windows and to
avoid accidental misuse there, Git always wants to write UTF-16 with a
BOM, and will refuse to read UTF-16 without it.

However, musl's iconv implementation writes UTF-16 without a BOM,
relying on the user to interpret it as big-endian. This causes t0028 and
the related functionality to fail, since Git won't read the file without
a BOM.

Add a Makefile and #define knob, ICONV_NEEDS_BOM, that can be set if the
iconv implementation has this behavior. When set, Git will write a BOM
manually for UTF-16 and UTF-32 and then force the data to be written in
UTF-16BE or UTF-32BE. We choose big-endian behavior here because the
tests use the raw "UTF-16" encoding, which will be big-endian when the
implementation requires this knob to be set.

Update the tests to detect this case and write test data with an added
BOM if necessary. Always write the BOM in the tests in big-endian
format, since all iconv implementations that omit a BOM must use
big-endian serialization according to the Unicode standard.

Preserve the existing behavior for systems which do not have this knob
enabled, since they may use optimized implementations, including
defaulting to the native endianness, to gain improved performance, which
can be significant with large checkouts.

Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
---
 Makefile                         |  6 ++++++
 t/t0028-working-tree-encoding.sh | 25 ++++++++++++++++++++++---
 utf8.c                           | 10 ++++++++++
 3 files changed, 38 insertions(+), 3 deletions(-)

Comments

Eric Sunshine Feb. 10, 2019, 1:45 a.m. UTC | #1
On Sat, Feb 9, 2019 at 3:08 PM brian m. carlson
<sandals@crustytoothpaste.net> wrote:
> [...]
> Add a Makefile and #define knob, ICONV_NEEDS_BOM, that can be set if the
> iconv implementation has this behavior. When set, Git will write a BOM
> manually for UTF-16 and UTF-32 and then force the data to be written in
> UTF-16BE or UTF-32BE. We choose big-endian behavior here because the
> tests use the raw "UTF-16" encoding, which will be big-endian when the
> implementation requires this knob to be set.

The name ICONV_NEEDS_BOM makes it sound as if we must feed a BOM
_into_ 'iconv', which is quite confusing since the actual intention is
that 'iconv' doesn't emit a BOM and we need to make up for the
deficiency. Using a name such as ICONV_OMITS_BOM or ICONV_NEGLECTS_BOM
makes it somewhat clearer that there is some deficiency with which we
need to deal.

> Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
> ---
> diff --git a/Makefile b/Makefile
> @@ -259,6 +259,9 @@ all::
> +# Define ICONV_NEEDS_BOM if your iconv implementation does not write a
> +# byte-order mark (BOM) when writing UTF-16 or UTF-32.

Not a big deal, but I wonder if it would be helpful to tack on "...,
in which case it outputs big-endian unconditionally." or something.

> diff --git a/t/t0028-working-tree-encoding.sh b/t/t0028-working-tree-encoding.sh
> @@ -6,6 +6,25 @@ test_description='working-tree-encoding conversion via gitattributes'
> +test_lazy_prereq NO_UTF16_BOM '
> +       test $(printf abc | iconv -f UTF-8 -t UTF-16 | wc -c) = 6
> +'
> +
> +test_lazy_prereq NO_UTF32_BOM '
> +       test $(printf abc | iconv -f UTF-8 -t UTF-32 | wc -c) = 12
> +'
> +
> +write_utf16 () {
> +       test_have_prereq NO_UTF16_BOM && printf '\xfe\xff'
> +       iconv -f UTF-8 -t UTF-16
> +
> +}

Stray blank line before the closing brace.

> +
> +write_utf32 () {
> +       test_have_prereq NO_UTF32_BOM && printf '\x00\x00\xfe\xff'
> +       iconv -f UTF-8 -t UTF-32
> +}

It's probably doesn't matter much with these two tiny functions, but I
was wondering if it would make sense to maintain the &&-chain, perhaps
like this:

    if test test_have_prereq NO_UTF32_BOM
    then
        printf '\x00\x00\xfe\xff'
    fi &&
    iconv -f UTF-8 -t UTF-32
Torsten Bögershausen Feb. 10, 2019, 8:04 a.m. UTC | #2
On Sat, Feb 09, 2019 at 08:08:01PM +0000, brian m. carlson wrote:
> When serializing UTF-16 (and UTF-32), there are three possible ways to
> write the stream. One can write the data with a BOM in either big-endian
> or little-endian format, or one can write the data without a BOM in
> big-endian format.
>
> Most systems' iconv implementations choose to write it with a BOM in
> some endianness, since this is the most foolproof, and it is resistant
> to misinterpretation on Windows, where UTF-16 and the little-endian
> serialization are very common. For compatibility with Windows and to
> avoid accidental misuse there, Git always wants to write UTF-16 with a
> BOM, and will refuse to read UTF-16 without it.
>
> However, musl's iconv implementation writes UTF-16 without a BOM,
> relying on the user to interpret it as big-endian. This causes t0028 and
> the related functionality to fail, since Git won't read the file without
> a BOM.
>
> Add a Makefile and #define knob, ICONV_NEEDS_BOM, that can be set if the
> iconv implementation has this behavior. When set, Git will write a BOM
> manually for UTF-16 and UTF-32 and then force the data to be written in
> UTF-16BE or UTF-32BE. We choose big-endian behavior here because the
> tests use the raw "UTF-16" encoding, which will be big-endian when the
> implementation requires this knob to be set.
>
> Update the tests to detect this case and write test data with an added
> BOM if necessary. Always write the BOM in the tests in big-endian
> format, since all iconv implementations that omit a BOM must use
> big-endian serialization according to the Unicode standard.
>
> Preserve the existing behavior for systems which do not have this knob
> enabled, since they may use optimized implementations, including
> defaulting to the native endianness, to gain improved performance, which
> can be significant with large checkouts.

Is the based on measurements on a real system ?

I think we agree that Git will write UTF-16 always as big endian with BOM,
following the tradition of iconv/libiconv.
If yes, we can reduce the lines of code/#idefs somewhat, have the knob always on,
and reduce the maintenance burden a little bit, giving a simpler patch.

What do you think ?


diff --git a/t/t0028-working-tree-encoding.sh b/t/t0028-working-tree-encoding.sh
index e58ecbfc44..ef19c98e67 100755
--- a/t/t0028-working-tree-encoding.sh
+++ b/t/t0028-working-tree-encoding.sh
@@ -13,7 +13,8 @@ test_expect_success 'setup test files' '
 	echo "*.utf16 text working-tree-encoding=utf-16" >.gitattributes &&
 	echo "*.utf16lebom text working-tree-encoding=UTF-16LE-BOM" >>.gitattributes &&
 	printf "$text" >test.utf8.raw &&
-	printf "$text" | iconv -f UTF-8 -t UTF-16 >test.utf16.raw &&
+	printf "\376\377"                         >test.utf16.raw &&
+	printf "$text" | iconv -f UTF-8 -t UTF-16BE >>test.utf16.raw &&
 	printf "$text" | iconv -f UTF-8 -t UTF-32 >test.utf32.raw &&
 	printf "\377\376"                         >test.utf16lebom.raw &&
 	printf "$text" | iconv -f UTF-8 -t UTF-32LE >>test.utf16lebom.raw &&
diff --git a/utf8.c b/utf8.c
index 83824dc2f4..d3731273be 100644
--- a/utf8.c
+++ b/utf8.c
@@ -564,7 +564,8 @@ char *reencode_string_len(const char *in, size_t insz,
 		bom_str = utf16_le_bom;
 		bom_len = sizeof(utf16_le_bom);
 		out_encoding = "UTF-16LE";
-	} else if (same_utf_encoding("UTF-16BE-BOM", out_encoding)) {
+	} else if (same_utf_encoding("UTF-16BE-BOM", out_encoding) ||
+		   same_utf_encoding("UTF-16", out_encoding)) {
 		bom_str = utf16_be_bom;
 		bom_len = sizeof(utf16_be_bom);
 		out_encoding = "UTF-16BE";


>
> Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
> ---
>  Makefile                         |  6 ++++++
>  t/t0028-working-tree-encoding.sh | 25 ++++++++++++++++++++++---
>  utf8.c                           | 10 ++++++++++
>  3 files changed, 38 insertions(+), 3 deletions(-)
>
> diff --git a/Makefile b/Makefile
> index 571160a2c4..b2a4765e5f 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -259,6 +259,9 @@ all::
>  # Define OLD_ICONV if your library has an old iconv(), where the second
>  # (input buffer pointer) parameter is declared with type (const char **).
>  #
> +# Define ICONV_NEEDS_BOM if your iconv implementation does not write a
> +# byte-order mark (BOM) when writing UTF-16 or UTF-32.
> +#
>  # Define NO_DEFLATE_BOUND if your zlib does not have deflateBound.
>  #
>  # Define NO_R_TO_GCC_LINKER if your gcc does not like "-R/path/lib"
> @@ -1415,6 +1418,9 @@ ifndef NO_ICONV
>  		EXTLIBS += $(ICONV_LINK) -liconv
>  	endif
>  endif
> +ifdef ICONV_NEEDS_BOM
> +	BASIC_CFLAGS += -DICONV_NEEDS_BOM
> +endif
>  ifdef NEEDS_LIBGEN
>  	EXTLIBS += -lgen
>  endif
> diff --git a/t/t0028-working-tree-encoding.sh b/t/t0028-working-tree-encoding.sh
> index e58ecbfc44..bfc4a9d4dd 100755
> --- a/t/t0028-working-tree-encoding.sh
> +++ b/t/t0028-working-tree-encoding.sh
> @@ -6,6 +6,25 @@ test_description='working-tree-encoding conversion via gitattributes'
>
>  GIT_TRACE_WORKING_TREE_ENCODING=1 && export GIT_TRACE_WORKING_TREE_ENCODING
>
> +test_lazy_prereq NO_UTF16_BOM '
> +	test $(printf abc | iconv -f UTF-8 -t UTF-16 | wc -c) = 6
> +'
> +
> +test_lazy_prereq NO_UTF32_BOM '
> +	test $(printf abc | iconv -f UTF-8 -t UTF-32 | wc -c) = 12
> +'
> +
> +write_utf16 () {
> +	test_have_prereq NO_UTF16_BOM && printf '\xfe\xff'
> +	iconv -f UTF-8 -t UTF-16
> +
> +}
> +
> +write_utf32 () {
> +	test_have_prereq NO_UTF32_BOM && printf '\x00\x00\xfe\xff'
> +	iconv -f UTF-8 -t UTF-32
> +}
> +
>  test_expect_success 'setup test files' '
>  	git config core.eol lf &&
>
> @@ -13,8 +32,8 @@ test_expect_success 'setup test files' '
>  	echo "*.utf16 text working-tree-encoding=utf-16" >.gitattributes &&
>  	echo "*.utf16lebom text working-tree-encoding=UTF-16LE-BOM" >>.gitattributes &&
>  	printf "$text" >test.utf8.raw &&
> -	printf "$text" | iconv -f UTF-8 -t UTF-16 >test.utf16.raw &&
> -	printf "$text" | iconv -f UTF-8 -t UTF-32 >test.utf32.raw &&
> +	printf "$text" | write_utf16 >test.utf16.raw &&
> +	printf "$text" | write_utf32 >test.utf32.raw &&
>  	printf "\377\376"                         >test.utf16lebom.raw &&
>  	printf "$text" | iconv -f UTF-8 -t UTF-32LE >>test.utf16lebom.raw &&
>
> @@ -223,7 +242,7 @@ test_expect_success ICONV_SHIFT_JIS 'check roundtrip encoding' '
>
>  	text="hallo there!\nroundtrip test here!" &&
>  	printf "$text" | iconv -f UTF-8 -t SHIFT-JIS >roundtrip.shift &&
> -	printf "$text" | iconv -f UTF-8 -t UTF-16 >roundtrip.utf16 &&
> +	printf "$text" | write_utf16 >roundtrip.utf16 &&
>  	echo "*.shift text working-tree-encoding=SHIFT-JIS" >>.gitattributes &&
>
>  	# SHIFT-JIS encoded files are round-trip checked by default...
> diff --git a/utf8.c b/utf8.c
> index 83824dc2f4..133199de0e 100644
> --- a/utf8.c
> +++ b/utf8.c
> @@ -568,6 +568,16 @@ char *reencode_string_len(const char *in, size_t insz,
>  		bom_str = utf16_be_bom;
>  		bom_len = sizeof(utf16_be_bom);
>  		out_encoding = "UTF-16BE";
> +#ifdef ICONV_NEEDS_BOM
> +	} else if (same_utf_encoding("UTF-16", out_encoding)) {
> +		bom_str = utf16_be_bom;
> +		bom_len = sizeof(utf16_be_bom);
> +		out_encoding = "UTF-16BE";
> +	} else if (same_utf_encoding("UTF-32", out_encoding)) {
> +		bom_str = utf32_be_bom;
> +		bom_len = sizeof(utf32_be_bom);
> +		out_encoding = "UTF-32BE";
> +#endif
>  	}
>
>  	conv = iconv_open(out_encoding, in_encoding);
brian m. carlson Feb. 10, 2019, 6:14 p.m. UTC | #3
On Sat, Feb 09, 2019 at 08:45:16PM -0500, Eric Sunshine wrote:
> On Sat, Feb 9, 2019 at 3:08 PM brian m. carlson
> <sandals@crustytoothpaste.net> wrote:
> > [...]
> > Add a Makefile and #define knob, ICONV_NEEDS_BOM, that can be set if the
> > iconv implementation has this behavior. When set, Git will write a BOM
> > manually for UTF-16 and UTF-32 and then force the data to be written in
> > UTF-16BE or UTF-32BE. We choose big-endian behavior here because the
> > tests use the raw "UTF-16" encoding, which will be big-endian when the
> > implementation requires this knob to be set.
> 
> The name ICONV_NEEDS_BOM makes it sound as if we must feed a BOM
> _into_ 'iconv', which is quite confusing since the actual intention is
> that 'iconv' doesn't emit a BOM and we need to make up for the
> deficiency. Using a name such as ICONV_OMITS_BOM or ICONV_NEGLECTS_BOM
> makes it somewhat clearer that there is some deficiency with which we
> need to deal.

That does sound like a better name.

> > Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
> > ---
> > diff --git a/Makefile b/Makefile
> > @@ -259,6 +259,9 @@ all::
> > +# Define ICONV_NEEDS_BOM if your iconv implementation does not write a
> > +# byte-order mark (BOM) when writing UTF-16 or UTF-32.
> 
> Not a big deal, but I wonder if it would be helpful to tack on "...,
> in which case it outputs big-endian unconditionally." or something.

Sure, I can add that.

> Stray blank line before the closing brace.

Will fix.

> It's probably doesn't matter much with these two tiny functions, but I
> was wondering if it would make sense to maintain the &&-chain, perhaps
> like this:
> 
>     if test test_have_prereq NO_UTF32_BOM
>     then
>         printf '\x00\x00\xfe\xff'
>     fi &&
>     iconv -f UTF-8 -t UTF-32

Yeah, that sounds like a good idea.
brian m. carlson Feb. 10, 2019, 6:55 p.m. UTC | #4
On Sun, Feb 10, 2019 at 08:04:13AM +0000, Torsten Bögershausen wrote:
> On Sat, Feb 09, 2019 at 08:08:01PM +0000, brian m. carlson wrote:
> > Preserve the existing behavior for systems which do not have this knob
> > enabled, since they may use optimized implementations, including
> > defaulting to the native endianness, to gain improved performance, which
> > can be significant with large checkouts.
> 
> Is the based on measurements on a real system ?

No, I haven't done any performance measurements. However, swapping bytes
is a (IIRC 1-cycle) instruction on x86, which would be executed for each
iteration of the loop. My intuition tells me that will be a significant
expense when there are a lot of files, but I can omit that phrase since
I haven't measured.

> I think we agree that Git will write UTF-16 always as big endian with BOM,
> following the tradition of iconv/libiconv.
> If yes, we can reduce the lines of code/#idefs somewhat, have the knob always on,
> and reduce the maintenance burden a little bit, giving a simpler patch.

No, I don't think it will. libiconv will always write big-endian, but
glibc has a separate iconv implementation which writes the native
endianness. (I believe FreeBSD's does the same thing as glibc's.) I
think it's useful for us to know that we can handle UTF-16 using the
system behavior where possible, since that's what the system is going to
produce.

> What do you think ?

While I like the simplicity of the approach, as I mentioned above, and I
did consider this originally, I'd rather test the behavior of the system
we're operating on, provided it's suitable for our needs.
Junio C Hamano Feb. 11, 2019, 5:14 p.m. UTC | #5
"brian m. carlson" <sandals@crustytoothpaste.net> writes:

> On Sun, Feb 10, 2019 at 08:04:13AM +0000, Torsten Bögershausen wrote:
>
>> I think we agree that Git will write UTF-16 always as big endian with BOM,
>> following the tradition of iconv/libiconv.
>> If yes, we can reduce the lines of code/#idefs somewhat, have the knob always on,
>> and reduce the maintenance burden a little bit, giving a simpler patch.
>
> No, I don't think it will. libiconv will always write big-endian, but
> glibc has a separate iconv implementation which writes the native
> endianness. (I believe FreeBSD's does the same thing as glibc's.) I
> think it's useful for us to know that we can handle UTF-16 using the
> system behavior where possible, since that's what the system is going to
> produce.
>
>> What do you think ?
>
> While I like the simplicity of the approach, as I mentioned above, and I
> did consider this originally, I'd rather test the behavior of the system
> we're operating on, provided it's suitable for our needs.

I see both sides of the argument, and each has its merit.

Let's go with the "follow the platform" and make sure the decision
is documented somewhere in the resulting code.

Thanks, all.

Patch
diff mbox series

diff --git a/Makefile b/Makefile
index 571160a2c4..b2a4765e5f 100644
--- a/Makefile
+++ b/Makefile
@@ -259,6 +259,9 @@  all::
 # Define OLD_ICONV if your library has an old iconv(), where the second
 # (input buffer pointer) parameter is declared with type (const char **).
 #
+# Define ICONV_NEEDS_BOM if your iconv implementation does not write a
+# byte-order mark (BOM) when writing UTF-16 or UTF-32.
+#
 # Define NO_DEFLATE_BOUND if your zlib does not have deflateBound.
 #
 # Define NO_R_TO_GCC_LINKER if your gcc does not like "-R/path/lib"
@@ -1415,6 +1418,9 @@  ifndef NO_ICONV
 		EXTLIBS += $(ICONV_LINK) -liconv
 	endif
 endif
+ifdef ICONV_NEEDS_BOM
+	BASIC_CFLAGS += -DICONV_NEEDS_BOM
+endif
 ifdef NEEDS_LIBGEN
 	EXTLIBS += -lgen
 endif
diff --git a/t/t0028-working-tree-encoding.sh b/t/t0028-working-tree-encoding.sh
index e58ecbfc44..bfc4a9d4dd 100755
--- a/t/t0028-working-tree-encoding.sh
+++ b/t/t0028-working-tree-encoding.sh
@@ -6,6 +6,25 @@  test_description='working-tree-encoding conversion via gitattributes'
 
 GIT_TRACE_WORKING_TREE_ENCODING=1 && export GIT_TRACE_WORKING_TREE_ENCODING
 
+test_lazy_prereq NO_UTF16_BOM '
+	test $(printf abc | iconv -f UTF-8 -t UTF-16 | wc -c) = 6
+'
+
+test_lazy_prereq NO_UTF32_BOM '
+	test $(printf abc | iconv -f UTF-8 -t UTF-32 | wc -c) = 12
+'
+
+write_utf16 () {
+	test_have_prereq NO_UTF16_BOM && printf '\xfe\xff'
+	iconv -f UTF-8 -t UTF-16
+
+}
+
+write_utf32 () {
+	test_have_prereq NO_UTF32_BOM && printf '\x00\x00\xfe\xff'
+	iconv -f UTF-8 -t UTF-32
+}
+
 test_expect_success 'setup test files' '
 	git config core.eol lf &&
 
@@ -13,8 +32,8 @@  test_expect_success 'setup test files' '
 	echo "*.utf16 text working-tree-encoding=utf-16" >.gitattributes &&
 	echo "*.utf16lebom text working-tree-encoding=UTF-16LE-BOM" >>.gitattributes &&
 	printf "$text" >test.utf8.raw &&
-	printf "$text" | iconv -f UTF-8 -t UTF-16 >test.utf16.raw &&
-	printf "$text" | iconv -f UTF-8 -t UTF-32 >test.utf32.raw &&
+	printf "$text" | write_utf16 >test.utf16.raw &&
+	printf "$text" | write_utf32 >test.utf32.raw &&
 	printf "\377\376"                         >test.utf16lebom.raw &&
 	printf "$text" | iconv -f UTF-8 -t UTF-32LE >>test.utf16lebom.raw &&
 
@@ -223,7 +242,7 @@  test_expect_success ICONV_SHIFT_JIS 'check roundtrip encoding' '
 
 	text="hallo there!\nroundtrip test here!" &&
 	printf "$text" | iconv -f UTF-8 -t SHIFT-JIS >roundtrip.shift &&
-	printf "$text" | iconv -f UTF-8 -t UTF-16 >roundtrip.utf16 &&
+	printf "$text" | write_utf16 >roundtrip.utf16 &&
 	echo "*.shift text working-tree-encoding=SHIFT-JIS" >>.gitattributes &&
 
 	# SHIFT-JIS encoded files are round-trip checked by default...
diff --git a/utf8.c b/utf8.c
index 83824dc2f4..133199de0e 100644
--- a/utf8.c
+++ b/utf8.c
@@ -568,6 +568,16 @@  char *reencode_string_len(const char *in, size_t insz,
 		bom_str = utf16_be_bom;
 		bom_len = sizeof(utf16_be_bom);
 		out_encoding = "UTF-16BE";
+#ifdef ICONV_NEEDS_BOM
+	} else if (same_utf_encoding("UTF-16", out_encoding)) {
+		bom_str = utf16_be_bom;
+		bom_len = sizeof(utf16_be_bom);
+		out_encoding = "UTF-16BE";
+	} else if (same_utf_encoding("UTF-32", out_encoding)) {
+		bom_str = utf32_be_bom;
+		bom_len = sizeof(utf32_be_bom);
+		out_encoding = "UTF-32BE";
+#endif
 	}
 
 	conv = iconv_open(out_encoding, in_encoding);