[Tarantool-patches] [PATCH luajit 0/2] Fix os.time() to use errno to detect errors

[Tarantool-patches] [PATCH luajit 0/2] Fix os.time() to use errno to detect errors

[Evgeniy Temirgaleev via Tarantool-patches]

This patchset fixes os.time() to make -1 time value 'legal' if no error occurs.

Branch: https://github.com/tarantool/luajit/tree/tmr_g/lj-1470-os-time-epoch-minus-1s
Related issues:
* https://github.com/tarantool/tarantool/issues/12480
* https://github.com/LuaJIT/LuaJIT/issues/1470

Mike Pall (2):
  Fix corner case of os.time().
  Make check in os.time() consistent.

 src/lib_os.c                                         |  3 ++-
 .../lj-1470-os-time-epoch-minus-1s.test.lua          | 12 ++++++++++++
 2 files changed, 14 insertions(+), 1 deletion(-)
 create mode 100644 test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua

-- 
2.49.0

[Tarantool-patches] [PATCH luajit 1/2] Fix corner case of os.time().

[Evgeniy Temirgaleev via Tarantool-patches]

From: Mike Pall <mike>

Thanks to Temir Galeev. #1470

(cherry picked from commit 72d2061ae2fa9a5fd45237943f9982baf59435ec)

This patch fixes invalid os.time(t) behavior, when it handles -1 time value
as a fail of the libc mktime() regardless of the errno value.

Temir Galeev:
* added the description and the test for the patch

Part of tarantool/tarantool#12480
---
 src/lib_os.c                                         |  3 ++-
 .../lj-1470-os-time-epoch-minus-1s.test.lua          | 12 ++++++++++++
 2 files changed, 14 insertions(+), 1 deletion(-)
 create mode 100644 test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua

diff --git a/src/lib_os.c b/src/lib_os.c
index ffbc3fdc..f6841357 100644
--- a/src/lib_os.c
+++ b/src/lib_os.c
@@ -242,9 +242,10 @@ LJLIB_CF(os_time)
     ts.tm_mon = getfield(L, "month", -1) - 1;
     ts.tm_year = getfield(L, "year", -1) - 1900;
     ts.tm_isdst = getboolfield(L, "isdst");
+    errno = 0;
     t = mktime(&ts);
   }
-  if (t == (time_t)(-1))
+  if (t == (time_t)(-1) && errno != 0)
     lua_pushnil(L);
   else
     lua_pushnumber(L, (lua_Number)t);
diff --git a/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
new file mode 100644
index 00000000..6b5703e4
--- /dev/null
+++ b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
@@ -0,0 +1,12 @@
+local tap = require('tap')
+
+-- The test file demonstrates os.time() fail to return -1 time value.
+-- See also: https://github.com/LuaJIT/LuaJIT/issues/1470.
+local test = tap.test('lj-1470-os-time-epoch-minus-1s')
+
+test:plan(1)
+
+local minus_1s_time = os.date("*t", -1)
+test:is(os.time(minus_1s_time), -1, 'correct os.time()')
+
+test:done(true)
-- 
2.49.0

Re: [Tarantool-patches] [PATCH luajit 1/2] Fix corner case of os.time().

[Sergey Kaplun via Tarantool-patches]

Hi, Evgeniy!
Thanks for the patch!
LGTM, with minor nits below.

On 29.06.26, Evgeniy Temirgaleev wrote:
> From: Mike Pall <mike>
> 
> Thanks to Temir Galeev. #1470

Please strip the issue number to avoid overmentioning the ticket on
GitHub. (The Issue can be easily found by the cherry-picked commit,
though). This helps to avoid overmentioning during the review and
applying the patch to long-term branches.

> 
> (cherry picked from commit 72d2061ae2fa9a5fd45237943f9982baf59435ec)
> 
> This patch fixes invalid os.time(t) behavior, when it handles -1 time value

Typo: s/, when/ when/
Typo: s/-1/the -1/

> as a fail of the libc mktime() regardless of the errno value.
> 
> Temir Galeev:
> * added the description and the test for the patch
> 
> Part of tarantool/tarantool#12480
> ---
>  src/lib_os.c                                         |  3 ++-
>  .../lj-1470-os-time-epoch-minus-1s.test.lua          | 12 ++++++++++++
>  2 files changed, 14 insertions(+), 1 deletion(-)
>  create mode 100644 test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> 
> diff --git a/src/lib_os.c b/src/lib_os.c
> index ffbc3fdc..f6841357 100644
> --- a/src/lib_os.c
> +++ b/src/lib_os.c
> @@ -242,9 +242,10 @@ LJLIB_CF(os_time)
>      ts.tm_mon = getfield(L, "month", -1) - 1;
>      ts.tm_year = getfield(L, "year", -1) - 1900;
>      ts.tm_isdst = getboolfield(L, "isdst");
> +    errno = 0;
>      t = mktime(&ts);
>    }
> -  if (t == (time_t)(-1))
> +  if (t == (time_t)(-1) && errno != 0)
>      lua_pushnil(L);
>    else
>      lua_pushnumber(L, (lua_Number)t);
> diff --git a/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> new file mode 100644
> index 00000000..6b5703e4
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> @@ -0,0 +1,12 @@
> +local tap = require('tap')
> +
> +-- The test file demonstrates os.time() fail to return -1 time value.

Side note: It looks like the patch is outdated compared to the branch.

> +-- See also: https://github.com/LuaJIT/LuaJIT/issues/1470.
> +local test = tap.test('lj-1470-os-time-epoch-minus-1s')
> +
> +test:plan(1)
> +
> +local minus_1s_time = os.date("*t", -1)

Nit: We use everywhere in Lua code single quotes if possible (see other
tests).

> +test:is(os.time(minus_1s_time), -1, 'correct os.time()')
> +
> +test:done(true)
> -- 
> 2.49.0
> 

-- 
Best regards,
Sergey Kaplun

Re: [Tarantool-patches] [PATCH luajit 1/2] Fix corner case of os.time().

[Evgeniy Temirgaleev via Tarantool-patches]

[-- Attachment #1: Type: text/plain, Size: 3575 bytes --]

Hi, Sergey!
Thanks for the review!
Fixed your comments and force-pushed the branch.

> 
> From: Sergey Kaplun <skaplun@tarantool.org>
> To: Evgeniy Temirgaleev <e.temirgaleev@tarantool.org>
> Cc: Sergey Bronnikov <sergeyb@tarantool.org>, tarantool-patches@dev.tarantool.org
> 
> Date: Tuesday, June 30, 2026 1:06 PM +03:00
> Hi, Evgeniy!
> Thanks for the patch!
> LGTM, with minor nits below.
> 
> On 29.06.26, Evgeniy Temirgaleev wrote:
> > From: Mike Pall <mike>
> >
> > Thanks to Temir Galeev. #1470
> 

> 
> 
> Please strip the issue number to avoid overmentioning the ticket on
> GitHub. (The Issue can be easily found by the cherry-picked commit,
> though). This helps to avoid overmentioning during the review and
> applying the patch to long-term branches.
> 

Removed.

> 
> 
> >
> > (cherry picked from commit 72d2061ae2fa9a5fd45237943f9982baf59435ec)
> >
> > This patch fixes invalid os.time(t) behavior, when it handles -1 time
> value
> 
> Typo: s/, when/ when/
> Typo: s/-1/the -1/
> 

Fixed.

> 
> 
> > as a fail of the libc mktime() regardless of the errno value.
> >
> > Temir Galeev:
> > * added the description and the test for the patch
> >
> > Part of tarantool/tarantool#12480
> > ---
> > src/lib_os.c | 3 ++-
> > .../lj-1470-os-time-epoch-minus-1s.test.lua | 12 ++++++++++++
> > 2 files changed, 14 insertions(+), 1 deletion(-)
> > create mode 100644
> test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> >
> > diff --git a/src/lib_os.c b/src/lib_os.c
> > index ffbc3fdc..f6841357 100644
> > --- a/src/lib_os.c
> > +++ b/src/lib_os.c
> > @@ -242,9 +242,10 @@ LJLIB_CF(os_time)
> > ts.tm_mon = getfield(L, "month", -1) - 1;
> > ts.tm_year = getfield(L, "year", -1) - 1900;
> > ts.tm_isdst = getboolfield(L, "isdst");
> > + errno = 0;
> > t = mktime(&ts);
> > }
> > - if (t == (time_t)(-1))
> > + if (t == (time_t)(-1) && errno != 0)
> > lua_pushnil(L);
> > else
> > lua_pushnumber(L, (lua_Number)t);
> > diff --git
> a/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> > new file mode 100644
> > index 00000000..6b5703e4
> > --- /dev/null
> > +++ b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> > @@ -0,0 +1,12 @@
> > +local tap = require('tap')
> > +
> > +-- The test file demonstrates os.time() fail to return -1 time value.
> 
> Side note: It looks like the patch is outdated compared to the branch.
> 

Yes, thanks!

> 
> 
> > +-- See also: https://github.com/LuaJIT/LuaJIT/issues/1470.
> > +local test = tap.test('lj-1470-os-time-epoch-minus-1s')
> > +
> > +test:plan(1)
> > +
> > +local minus_1s_time = os.date("*t", -1)
> 
> Nit: We use everywhere in Lua code single quotes if possible (see other
> tests).
> 

Fixed.
======================================
diff --git a/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
index aa913607..dfe1ee11 100644
--- a/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
+++ b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
@@ -7,7 +7,7 @@ local test = tap.test('lj-1470-os-time-epoch-minus-1s')

test:plan(1)

-local minus_1s_time = os.date("*t", -1)
+local minus_1s_time = os.date('*t', -1)
test:is(os.time(minus_1s_time), -1, 'correct os.time()')

test:done(true)
======================================

> 
> 
> > +test:is(os.time(minus_1s_time), -1, 'correct os.time()')
> > +
> > +test:done(true)
> > --
> > 2.49.0
> >
> 
> --
> Best regards,
> Sergey Kaplun
> 

--
Best regards,
Evgeniy Temirgaleev

[-- Attachment #2: Type: text/html, Size: 7737 bytes --]

Re: [Tarantool-patches] [PATCH luajit 1/2] Fix corner case of os.time().

[Sergey Bronnikov via Tarantool-patches]

[-- Attachment #1: Type: text/plain, Size: 1944 bytes --]

Hi, Evgeniy!

Thanks for the patch! LGTM

Sergey

On 6/29/26 20:21, Evgeniy Temirgaleev wrote:
> From: Mike Pall <mike>
>
> Thanks to Temir Galeev. #1470
>
> (cherry picked from commit 72d2061ae2fa9a5fd45237943f9982baf59435ec)
>
> This patch fixes invalid os.time(t) behavior, when it handles -1 time value
> as a fail of the libc mktime() regardless of the errno value.
>
> Temir Galeev:
> * added the description and the test for the patch
>
> Part of tarantool/tarantool#12480
> ---
>   src/lib_os.c                                         |  3 ++-
>   .../lj-1470-os-time-epoch-minus-1s.test.lua          | 12 ++++++++++++
>   2 files changed, 14 insertions(+), 1 deletion(-)
>   create mode 100644 test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
>
> diff --git a/src/lib_os.c b/src/lib_os.c
> index ffbc3fdc..f6841357 100644
> --- a/src/lib_os.c
> +++ b/src/lib_os.c
> @@ -242,9 +242,10 @@ LJLIB_CF(os_time)
>       ts.tm_mon = getfield(L, "month", -1) - 1;
>       ts.tm_year = getfield(L, "year", -1) - 1900;
>       ts.tm_isdst = getboolfield(L, "isdst");
> +    errno = 0;
>       t = mktime(&ts);
>     }
> -  if (t == (time_t)(-1))
> +  if (t == (time_t)(-1) && errno != 0)
>       lua_pushnil(L);
>     else
>       lua_pushnumber(L, (lua_Number)t);
> diff --git a/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> new file mode 100644
> index 00000000..6b5703e4
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1470-os-time-epoch-minus-1s.test.lua
> @@ -0,0 +1,12 @@
> +local tap = require('tap')
> +
> +-- The test file demonstrates os.time() fail to return -1 time value.
> +-- See also:https://github.com/LuaJIT/LuaJIT/issues/1470.
> +local test = tap.test('lj-1470-os-time-epoch-minus-1s')
> +
> +test:plan(1)
> +
> +local minus_1s_time = os.date("*t", -1)
> +test:is(os.time(minus_1s_time), -1, 'correct os.time()')
> +
> +test:done(true)

[-- Attachment #2: Type: text/html, Size: 2519 bytes --]

[Tarantool-patches] [PATCH luajit 2/2] Make check in os.time() consistent.

[Evgeniy Temirgaleev via Tarantool-patches]

From: Mike Pall <mike>

Thanks to Temir Galeev. #1470

(cherry picked from commit 295d45fb26de56498782c94594f31b97aef744ef)

This patch prevents os.time() to return nil, when machine time is -1
and some error occurs before os.time() call. In fact, this is
a theoretical case, so no tests were added.

Temir Galeev:
* added the description for the patch

Part of tarantool/tarantool#12480
---
 src/lib_os.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/lib_os.c b/src/lib_os.c
index f6841357..d4054eae 100644
--- a/src/lib_os.c
+++ b/src/lib_os.c
@@ -229,6 +229,7 @@ LJLIB_CF(os_date)
 LJLIB_CF(os_time)
 {
   time_t t;
+  errno = 0;
   if (lua_isnoneornil(L, 1)) {  /* called without args? */
     t = time(NULL);  /* get current time */
   } else {
@@ -242,7 +243,6 @@ LJLIB_CF(os_time)
     ts.tm_mon = getfield(L, "month", -1) - 1;
     ts.tm_year = getfield(L, "year", -1) - 1900;
     ts.tm_isdst = getboolfield(L, "isdst");
-    errno = 0;
     t = mktime(&ts);
   }
   if (t == (time_t)(-1) && errno != 0)
-- 
2.49.0

Re: [Tarantool-patches] [PATCH luajit 2/2] Make check in os.time() consistent.

[Sergey Kaplun via Tarantool-patches]

Hi, Evgeniy!
Thanks for the patch!
LGTM, with a few nits regarding the commit message.

On 29.06.26, Evgeniy Temirgaleev wrote:
> From: Mike Pall <mike>
> 
> Thanks to Temir Galeev. #1470

Please strip the issue number here. See the rationale in the response to
the previous patch.

> 
> (cherry picked from commit 295d45fb26de56498782c94594f31b97aef744ef)
> 
> This patch prevents os.time() to return nil, when machine time is -1

Typo: s/to return nil/from returning nil/

> and some error occurs before os.time() call. In fact, this is

Typo: s/os.time() call/the os.time() call/

> a theoretical case, so no tests were added.
> 
> Temir Galeev:
> * added the description for the patch
> 
> Part of tarantool/tarantool#12480
> ---

<snipped>

-- 
Best regards,
Sergey Kaplun

Re: [Tarantool-patches] [PATCH luajit 2/2] Make check in os.time() consistent.

[Evgeniy Temirgaleev via Tarantool-patches]

[-- Attachment #1: Type: text/plain, Size: 1254 bytes --]

Hi, Sergey! Thanks for the review!
Fixed your comments and force-pushed the branch.

> 
> From: Sergey Kaplun <skaplun@tarantool.org>
> To: Evgeniy Temirgaleev <e.temirgaleev@tarantool.org>
> Cc: Sergey Bronnikov <sergeyb@tarantool.org>, tarantool-patches@dev.tarantool.org
> 
> Date: Tuesday, June 30, 2026 1:09 PM +03:00
> Hi, Evgeniy!
> Thanks for the patch!
> LGTM, with a few nits regarding the commit message.
> 
> On 29.06.26, Evgeniy Temirgaleev wrote:
> > From: Mike Pall <mike>
> >
> > Thanks to Temir Galeev. #1470
> 
> Please strip the issue number here. See the rationale in the response to
> the previous patch.
> 

Removed.

> 
> 
> >
> > (cherry picked from commit 295d45fb26de56498782c94594f31b97aef744ef)
> >
> > This patch prevents os.time() to return nil, when machine time is -1
> 
> Typo: s/to return nil/from returning nil/
> 

Fixed.

> 
> 
> > and some error occurs before os.time() call. In fact, this is
> 
> Typo: s/os.time() call/the os.time() call/
> 

Fixed.

> 
> 
> > a theoretical case, so no tests were added.
> >
> > Temir Galeev:
> > * added the description for the patch
> >
> > Part of tarantool/tarantool#12480
> > ---
> 
> <snipped>
> 
> --
> Best regards,
> Sergey Kaplun
> 

--
Best regards,
Evgeniy Temirgaleev

[-- Attachment #2: Type: text/html, Size: 2966 bytes --]

Re: [Tarantool-patches] [PATCH luajit 2/2] Make check in os.time() consistent.

[Sergey Bronnikov via Tarantool-patches]

[-- Attachment #1: Type: text/plain, Size: 1217 bytes --]

Hi, Evgeniy!

Thanks for the patch! LGTM

Sergey

On 6/29/26 20:21, Evgeniy Temirgaleev wrote:
> From: Mike Pall <mike>
>
> Thanks to Temir Galeev. #1470
>
> (cherry picked from commit 295d45fb26de56498782c94594f31b97aef744ef)
>
> This patch prevents os.time() to return nil, when machine time is -1
> and some error occurs before os.time() call. In fact, this is
> a theoretical case, so no tests were added.
>
> Temir Galeev:
> * added the description for the patch
>
> Part of tarantool/tarantool#12480
> ---
>   src/lib_os.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/src/lib_os.c b/src/lib_os.c
> index f6841357..d4054eae 100644
> --- a/src/lib_os.c
> +++ b/src/lib_os.c
> @@ -229,6 +229,7 @@ LJLIB_CF(os_date)
>   LJLIB_CF(os_time)
>   {
>     time_t t;
> +  errno = 0;
>     if (lua_isnoneornil(L, 1)) {  /* called without args? */
>       t = time(NULL);  /* get current time */
>     } else {
> @@ -242,7 +243,6 @@ LJLIB_CF(os_time)
>       ts.tm_mon = getfield(L, "month", -1) - 1;
>       ts.tm_year = getfield(L, "year", -1) - 1900;
>       ts.tm_isdst = getboolfield(L, "isdst");
> -    errno = 0;
>       t = mktime(&ts);
>     }
>     if (t == (time_t)(-1) && errno != 0)

[-- Attachment #2: Type: text/html, Size: 1662 bytes --]

Re: [Tarantool-patches] [PATCH luajit 0/2] Fix os.time() to use errno to detect errors

[Sergey Kaplun via Tarantool-patches]

Evgeniy,
Thanks for the patch-set and fixes!
LGTM!

-- 
Best regards,
Sergey Kaplun