[musl] strptime_l implementation

[musl] strptime_l implementation

[Javier Steinaker]

[-- Attachment #1: Type: text/plain, Size: 1019 bytes --]

Hello everybody,

The Haiku project (an alternative OS) switched to using musl a few weeks
ago. Now I'm only an occasional contributor, but I hit an use case where
having strptime_l would be desired/useful (parsing web cookies, which are
always in english, independently of the locale selected by the user). Since
nl_langinfo_l is already in place, I figured out it wouldn't be so
difficult to move the current code to an internal function, and then have
strptime and strptime_l getting the locale from the system in the first
case and as an argument in the second, and call that code.

My question is: do you have plans to have strptime_l implemented? Would you
be interested in merging if someone does? Would it break the ABI or
something? I'm asking because it made more sense to me to do this upstream
and then don't having to maintain a separate version if it was useful here.
Otherwise, we will just find our way in the Haiku code, via implementing
strptime_l or with another solution.

Thanks in advance,
Javier

[-- Attachment #2: Type: text/html, Size: 1185 bytes --]

Re: [musl] strptime_l implementation

[Markus Wichmann]

On Thu, Sep 08, 2022 at 05:36:34AM +1000, Javier Steinaker wrote:
> Hello everybody,
>
> The Haiku project (an alternative OS) switched to using musl a few weeks
> ago. Now I'm only an occasional contributor, but I hit an use case where
> having strptime_l would be desired/useful (parsing web cookies, which are
> always in english, independently of the locale selected by the user). Since
> nl_langinfo_l is already in place, I figured out it wouldn't be so
> difficult to move the current code to an internal function, and then have
> strptime and strptime_l getting the locale from the system in the first
> case and as an argument in the second, and call that code.
>
> My question is: do you have plans to have strptime_l implemented? Would you
> be interested in merging if someone does? Would it break the ABI or
> something? I'm asking because it made more sense to me to do this upstream
> and then don't having to maintain a separate version if it was useful here.
> Otherwise, we will just find our way in the Haiku code, via implementing
> strptime_l or with another solution.
>
> Thanks in advance,
> Javier

Standard requirements for inclusion are absolutely any kind of
standardization effort or agreement among libcs. Otherwise we run the
risk that later on, a different version is standardized, and then we
cannot change our version without breaking ABI.

No, at this point, adding strptime_l would not break ABI, but if we had
to change the interface later on, then it would. So better to only add
it once everything is coordinated. And if I search for the proposed name
of the function, I find a manpage for Illumos, one for Gnulib, and one
at mkssoftware.com. That last one claims that strptime_l() is conformant
to POSIX, but I failed to find it in POSIX, so I think that claim is
wrong.

So basically, that is one necessary condition I do not find fulfilled
here.

Ciao,
Markus

Re: [musl] strptime_l implementation

[Rich Felker]

On Thu, Sep 08, 2022 at 05:36:34AM +1000, Javier Steinaker wrote:
> Hello everybody,
> 
> The Haiku project (an alternative OS) switched to using musl a few weeks
> ago. Now I'm only an occasional contributor, but I hit an use case where
> having strptime_l would be desired/useful (parsing web cookies, which are
> always in english, independently of the locale selected by the user). Since
> nl_langinfo_l is already in place, I figured out it wouldn't be so
> difficult to move the current code to an internal function, and then have
> strptime and strptime_l getting the locale from the system in the first
> case and as an argument in the second, and call that code.
> 
> My question is: do you have plans to have strptime_l implemented? Would you
> be interested in merging if someone does? Would it break the ABI or
> something? I'm asking because it made more sense to me to do this upstream
> and then don't having to maintain a separate version if it was useful here.
> Otherwise, we will just find our way in the Haiku code, via implementing
> strptime_l or with another solution.

There are some number of nonstandard *_l functions, some of which have
been requested for addition. My request on these has been for someone
to do a survey of how many there actually are, and whether it makes
sense to add them all or some well-characterized subset of them. I
don't want to just inconsistently add one here and there, while
leaving others missing, and I don't want to get in a situation where
we feel obligated to add a bunch of dubious interfaces just by
precedent.

If you'd like to make such a list/count, I'm sure it would be
appreciated by others who have raised the topic before, and might end
up with the functions getting added.

Short of that, the portable way to do locale_t-parameterized calls to
functions that don't have their own *_l version is by calling
uselocale() before/after to save/swap and restore the original locale.
This operation costs essentially nothing and is a completely viable
way to do things.

Rich

Re: [musl] strptime_l implementation

[James Y Knight]

[-- Attachment #1: Type: text/plain, Size: 2338 bytes --]

Someone in 2016 compared POSIX, cygwin, and glibc's list of locale_t
functions: https://www.mail-archive.com/cygwin@cygwin.com/msg149338.html



On Wed, Sep 7, 2022 at 4:49 PM Rich Felker <dalias@libc.org> wrote:

> On Thu, Sep 08, 2022 at 05:36:34AM +1000, Javier Steinaker wrote:
> > Hello everybody,
> >
> > The Haiku project (an alternative OS) switched to using musl a few weeks
> > ago. Now I'm only an occasional contributor, but I hit an use case where
> > having strptime_l would be desired/useful (parsing web cookies, which are
> > always in english, independently of the locale selected by the user).
> Since
> > nl_langinfo_l is already in place, I figured out it wouldn't be so
> > difficult to move the current code to an internal function, and then have
> > strptime and strptime_l getting the locale from the system in the first
> > case and as an argument in the second, and call that code.
> >
> > My question is: do you have plans to have strptime_l implemented? Would
> you
> > be interested in merging if someone does? Would it break the ABI or
> > something? I'm asking because it made more sense to me to do this
> upstream
> > and then don't having to maintain a separate version if it was useful
> here.
> > Otherwise, we will just find our way in the Haiku code, via implementing
> > strptime_l or with another solution.
>
> There are some number of nonstandard *_l functions, some of which have
> been requested for addition. My request on these has been for someone
> to do a survey of how many there actually are, and whether it makes
> sense to add them all or some well-characterized subset of them. I
> don't want to just inconsistently add one here and there, while
> leaving others missing, and I don't want to get in a situation where
> we feel obligated to add a bunch of dubious interfaces just by
> precedent.
>
> If you'd like to make such a list/count, I'm sure it would be
> appreciated by others who have raised the topic before, and might end
> up with the functions getting added.
>
> Short of that, the portable way to do locale_t-parameterized calls to
> functions that don't have their own *_l version is by calling
> uselocale() before/after to save/swap and restore the original locale.
> This operation costs essentially nothing and is a completely viable
> way to do things.
>
> Rich
>

[-- Attachment #2: Type: text/html, Size: 2933 bytes --]

Re: [musl] strptime_l implementation

[Rich Felker]

On Wed, Sep 07, 2022 at 06:46:32PM -0400, James Y Knight wrote:
> Someone in 2016 compared POSIX, cygwin, and glibc's list of locale_t
> functions: https://www.mail-archive.com/cygwin@cygwin.com/msg149338.html

If I'm reading that correctly, is the full list of nonstandard
extensions is:

{is,to}ascii_l
{str,wcs}to{l,ll,ul,ull,d,f,ld}_l
wcsftime_l
strptime_l

and the ones we're lacking are:

{is,to}ascii_l
{str,wcs}to{l,ll,ul,ull}_l
wcsto{d,f,ld}_l
strptime_l

If this is actually complete, strptime_l is the only one that would
not just be a pure thunk to throw away the locale_t arg.

From the above list, {is,to}ascii_l seem like junk I'd be inclined to
omit. They're _l versions of deprecated functions that should not be
used.

I'm not sure about {str,wcs}to{l,ll,ul,ull}_l. My leaning is that it's
misguided for apps to be using them. Their behavior is not locale
dependent per the standard, except possibly for admiting additional
locale-specific sequences to be interpreted as numbers. Perhaps that's
the rationale: being able to suppress that allowance by making a
LC_NUMERIC=C locale object and ensure only standard sequences are
supported. So these seem like a maybe.

wcsto{d,f,ld}_l should clearly be added for consistency with
strto{d,f,ld}_l which we already have.

strptime_l then seems like it can just be considered on its own
merits.


> On Wed, Sep 7, 2022 at 4:49 PM Rich Felker <dalias@libc.org> wrote:
> 
> > On Thu, Sep 08, 2022 at 05:36:34AM +1000, Javier Steinaker wrote:
> > > Hello everybody,
> > >
> > > The Haiku project (an alternative OS) switched to using musl a few weeks
> > > ago. Now I'm only an occasional contributor, but I hit an use case where
> > > having strptime_l would be desired/useful (parsing web cookies, which are
> > > always in english, independently of the locale selected by the user).
> > Since
> > > nl_langinfo_l is already in place, I figured out it wouldn't be so
> > > difficult to move the current code to an internal function, and then have
> > > strptime and strptime_l getting the locale from the system in the first
> > > case and as an argument in the second, and call that code.
> > >
> > > My question is: do you have plans to have strptime_l implemented? Would
> > you
> > > be interested in merging if someone does? Would it break the ABI or
> > > something? I'm asking because it made more sense to me to do this
> > upstream
> > > and then don't having to maintain a separate version if it was useful
> > here.
> > > Otherwise, we will just find our way in the Haiku code, via implementing
> > > strptime_l or with another solution.
> >
> > There are some number of nonstandard *_l functions, some of which have
> > been requested for addition. My request on these has been for someone
> > to do a survey of how many there actually are, and whether it makes
> > sense to add them all or some well-characterized subset of them. I
> > don't want to just inconsistently add one here and there, while
> > leaving others missing, and I don't want to get in a situation where
> > we feel obligated to add a bunch of dubious interfaces just by
> > precedent.
> >
> > If you'd like to make such a list/count, I'm sure it would be
> > appreciated by others who have raised the topic before, and might end
> > up with the functions getting added.
> >
> > Short of that, the portable way to do locale_t-parameterized calls to
> > functions that don't have their own *_l version is by calling
> > uselocale() before/after to save/swap and restore the original locale.
> > This operation costs essentially nothing and is a completely viable
> > way to do things.
> >
> > Rich
> >

Re: [musl] strptime_l implementation

[Javier Steinaker]

[-- Attachment #1: Type: text/plain, Size: 4613 bytes --]

Thank you all for your kind responses.

It seems that the uselocale() approach that Rich suggests is the easiest
thing to do for us for now. However, it's nice to see strptime_l being
considered.

In that case and just out of curiosity since my coding level it's far from
the standards seen here, could it be possible to make an internal function
that both strptime and strptime_l call providing the locale needed? ABI
shouldn't break as long as the interface for calling strptime remains the
same, am I right?

Thank you again for all the tips and information provided, I really
appreciate it.

El jue., 8 de septiembre de 2022 10:03, Rich Felker <dalias@libc.org>
escribió:

> On Wed, Sep 07, 2022 at 06:46:32PM -0400, James Y Knight wrote:
> > Someone in 2016 compared POSIX, cygwin, and glibc's list of locale_t
> > functions: https://www.mail-archive.com/cygwin@cygwin.com/msg149338.html
>
> If I'm reading that correctly, is the full list of nonstandard
> extensions is:
>
> {is,to}ascii_l
> {str,wcs}to{l,ll,ul,ull,d,f,ld}_l
> wcsftime_l
> strptime_l
>
> and the ones we're lacking are:
>
> {is,to}ascii_l
> {str,wcs}to{l,ll,ul,ull}_l
> wcsto{d,f,ld}_l
> strptime_l
>
> If this is actually complete, strptime_l is the only one that would
> not just be a pure thunk to throw away the locale_t arg.
>
> From the above list, {is,to}ascii_l seem like junk I'd be inclined to
> omit. They're _l versions of deprecated functions that should not be
> used.
>
> I'm not sure about {str,wcs}to{l,ll,ul,ull}_l. My leaning is that it's
> misguided for apps to be using them. Their behavior is not locale
> dependent per the standard, except possibly for admiting additional
> locale-specific sequences to be interpreted as numbers. Perhaps that's
> the rationale: being able to suppress that allowance by making a
> LC_NUMERIC=C locale object and ensure only standard sequences are
> supported. So these seem like a maybe.
>
> wcsto{d,f,ld}_l should clearly be added for consistency with
> strto{d,f,ld}_l which we already have.
>
> strptime_l then seems like it can just be considered on its own
> merits.
>
>
> > On Wed, Sep 7, 2022 at 4:49 PM Rich Felker <dalias@libc.org> wrote:
> >
> > > On Thu, Sep 08, 2022 at 05:36:34AM +1000, Javier Steinaker wrote:
> > > > Hello everybody,
> > > >
> > > > The Haiku project (an alternative OS) switched to using musl a few
> weeks
> > > > ago. Now I'm only an occasional contributor, but I hit an use case
> where
> > > > having strptime_l would be desired/useful (parsing web cookies,
> which are
> > > > always in english, independently of the locale selected by the user).
> > > Since
> > > > nl_langinfo_l is already in place, I figured out it wouldn't be so
> > > > difficult to move the current code to an internal function, and then
> have
> > > > strptime and strptime_l getting the locale from the system in the
> first
> > > > case and as an argument in the second, and call that code.
> > > >
> > > > My question is: do you have plans to have strptime_l implemented?
> Would
> > > you
> > > > be interested in merging if someone does? Would it break the ABI or
> > > > something? I'm asking because it made more sense to me to do this
> > > upstream
> > > > and then don't having to maintain a separate version if it was useful
> > > here.
> > > > Otherwise, we will just find our way in the Haiku code, via
> implementing
> > > > strptime_l or with another solution.
> > >
> > > There are some number of nonstandard *_l functions, some of which have
> > > been requested for addition. My request on these has been for someone
> > > to do a survey of how many there actually are, and whether it makes
> > > sense to add them all or some well-characterized subset of them. I
> > > don't want to just inconsistently add one here and there, while
> > > leaving others missing, and I don't want to get in a situation where
> > > we feel obligated to add a bunch of dubious interfaces just by
> > > precedent.
> > >
> > > If you'd like to make such a list/count, I'm sure it would be
> > > appreciated by others who have raised the topic before, and might end
> > > up with the functions getting added.
> > >
> > > Short of that, the portable way to do locale_t-parameterized calls to
> > > functions that don't have their own *_l version is by calling
> > > uselocale() before/after to save/swap and restore the original locale.
> > > This operation costs essentially nothing and is a completely viable
> > > way to do things.
> > >
> > > Rich
> > >
>

[-- Attachment #2: Type: text/html, Size: 5755 bytes --]