mbox series

[v2,0/4] builtin add -p: hopefully final readkey fixes

Message ID 20220309110325.36917-1-phillip.wood123@gmail.com (mailing list archive)
Headers show
Series builtin add -p: hopefully final readkey fixes | expand

Message

Phillip Wood March 9, 2022, 11:03 a.m. UTC
From: Phillip Wood <phillip.wood@dunelm.org.uk>

Thanks to Ramsay and Ævar for their comments on V1.
Changes since V1:
 * Patch 1
   - use an enum for save_term()'s flags (suggested by Ævar)
   - fixed argument order in the windows code (thanks to Ramsay)
 * Patch 2
   - fixed a typo in a comment (thanks to Ramsay)
 * Patch 4
   - stopped duplicating the strings returned by gettext() (suggested by
     Ævar)
   - reworked error message handling in the signal handler to add an
     "error: " prefix (suggested by Ævar)
   - tweaked the background resume error message

V1 Cover Letter:

Fix the remaining issues that I'm aware of when using the built in
"add -p" with interactive.singlekey that are stopping it from being
merged to master. The first three patches make sure that we call
tcsetattr() and the same file descriptor that we use for read() and
work around poll() being broken when reading from terminals on
macos. The final patch is more of an improvement rather than a bug fix
(the same issue already exists in the perl version) and could proceed
separately.

Unfortunately these patches conflict with
'cb/save-term-across-editor-invocation' as well as the textual
conflicts there is a semantic conflict as the argument to save_term()
is changed so the code in editor.c will need updating.

These patches are based on 'pw/single-key-interactive'

Phillip Wood (4):
  terminal: use flags for save_term()
  terminal: don't assume stdin is /dev/tty
  terminal: work around macos poll() bug
  terminal: restore settings on SIGTSTP

 compat/terminal.c | 206 +++++++++++++++++++++++++++++++++++++++-------
 compat/terminal.h |   9 +-
 2 files changed, 186 insertions(+), 29 deletions(-)

Range-diff against v1:
1:  6961a4e0dc ! 1:  fac5a0f5b7 terminal: use flags for save_term()
    @@ Commit message
         terminal: use flags for save_term()
     
         The next commit will add another flag in addition to the existing
    -    full_duplex so change the function signature to take an unsigned flags
    +    full_duplex so change the function signature to take a flags
         argument. Also alter the functions that call save_term() so that they
         can pass flags down to it.
     
    +    The choice to use an enum for tho bitwise flags is because gdb will
    +    display the symbolic names of all the flags that are set rather than
    +    the integer value.
    +
         Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
     
      ## compat/terminal.c ##
    @@ compat/terminal.c: void restore_term(void)
      }
      
     -int save_term(int full_duplex)
    -+int save_term(unsigned flags)
    ++int save_term(enum save_term_flags flags)
      {
      	if (term_fd < 0)
      		term_fd = open("/dev/tty", O_RDWR);
    @@ compat/terminal.c: int save_term(int full_duplex)
      }
      
     -static int disable_bits(tcflag_t bits)
    -+static int disable_bits(unsigned flags, tcflag_t bits)
    ++static int disable_bits(enum save_term_flags flags, tcflag_t bits)
      {
      	struct termios t;
      
    @@ compat/terminal.c: static int disable_bits(tcflag_t bits)
      }
      
     -static int disable_echo(void)
    -+static int disable_echo(unsigned flags)
    ++static int disable_echo(enum save_term_flags flags)
      {
     -	return disable_bits(ECHO);
     +	return disable_bits(flags, ECHO);
      }
      
     -static int enable_non_canonical(void)
    -+static int enable_non_canonical(unsigned flags)
    ++static int enable_non_canonical(enum save_term_flags flags)
      {
     -	return disable_bits(ICANON | ECHO);
     +	return disable_bits(flags, ICANON | ECHO);
    @@ compat/terminal.c: void restore_term(void)
      }
      
     -int save_term(int full_duplex)
    -+int save_term(unsigned flags)
    ++int save_term(enum save_term_flags flags)
      {
      	hconin = CreateFileA("CONIN$", GENERIC_READ | GENERIC_WRITE,
      	    FILE_SHARE_READ, NULL, OPEN_EXISTING,
    @@ compat/terminal.c: int save_term(int full_duplex)
      }
      
     -static int disable_bits(DWORD bits)
    -+static int disable_bits(unsigned flags, DWORD bits)
    ++static int disable_bits(enum save_term_flags flags, DWORD bits)
      {
      	if (use_stty) {
      		struct child_process cp = CHILD_PROCESS_INIT;
    @@ compat/terminal.c: static int disable_bits(DWORD bits)
      }
      
     -static int disable_echo(void)
    -+static int disable_echo(unsigned flags)
    ++static int disable_echo(enum save_term_flags flags)
      {
     -	return disable_bits(ENABLE_ECHO_INPUT);
    -+	return disable_bits(ENABLE_ECHO_INPUT, flags);
    ++	return disable_bits(flags, ENABLE_ECHO_INPUT);
      }
      
     -static int enable_non_canonical(void)
    -+static int enable_non_canonical(unsigned flags)
    ++static int enable_non_canonical(enum save_term_flags flags)
      {
     -	return disable_bits(ENABLE_ECHO_INPUT | ENABLE_LINE_INPUT | ENABLE_PROCESSED_INPUT);
    -+	return disable_bits(ENABLE_ECHO_INPUT | ENABLE_LINE_INPUT | ENABLE_PROCESSED_INPUT,
    -+			    flags);
    ++	return disable_bits(flags,
    ++			    ENABLE_ECHO_INPUT | ENABLE_LINE_INPUT | ENABLE_PROCESSED_INPUT);
      }
      
      /*
    @@ compat/terminal.c: int read_key_without_echo(struct strbuf *buf)
      #else
      
     -int save_term(int full_duplex)
    -+int save_term(unsigned flags)
    ++int save_term(enum save_term_flags flags)
      {
     -	/* full_duplex == 1, but no support available */
     -	return -full_duplex;
    @@ compat/terminal.h
      #ifndef COMPAT_TERMINAL_H
      #define COMPAT_TERMINAL_H
      
    -+/* Save input and output settings */
    -+#define SAVE_TERM_DUPLEX (1u << 0)
    ++enum save_term_flags {
    ++	/* Save input and output settings */
    ++	SAVE_TERM_DUPLEX = 1 << 0,
    ++};
     +
      /*
       * Save the terminal attributes so they can be restored later by a
2:  45315087d4 ! 2:  bf29460ec6 terminal: don't assume stdin is /dev/tty
    @@ Commit message
     
      ## compat/terminal.c ##
     @@ compat/terminal.c: void restore_term(void)
    - int save_term(unsigned flags)
    + int save_term(enum save_term_flags flags)
      {
      	if (term_fd < 0)
     -		term_fd = open("/dev/tty", O_RDWR);
    @@ compat/terminal.c: int read_key_without_echo(struct strbuf *buf)
     
      ## compat/terminal.h ##
     @@
    + enum save_term_flags {
    + 	/* Save input and output settings */
    + 	SAVE_TERM_DUPLEX = 1 << 0,
    ++	/* Save stdin rather than /dev/tty (fails if stdin is not a terminal) */
    ++	SAVE_TERM_STDIN  = 1 << 1,
    + };
      
    - /* Save input and output settings */
    - #define SAVE_TERM_DUPLEX (1u << 0)
    --
    -+/* Save stdin rather than /dev/tty (fails is stdin is not a terminal) */
    -+#define SAVE_TERM_STDIN  (1u << 1)
      /*
    -  * Save the terminal attributes so they can be restored later by a
    -  * call to restore_term(). Note that every successful call to
4:  62703be209 ! 4:  dd0e1fabb1 terminal: restore settings on SIGTSTP
    @@ compat/terminal.c: static void restore_term_on_signal(int sig)
      static int term_fd = -1;
      static struct termios old_term;
      
    -+static char *background_resume_msg;
    -+static char *restore_error_msg;
    ++static const char *background_resume_msg;
    ++static const char *restore_error_msg;
     +static volatile sig_atomic_t ttou_received;
     +
    -+static void write_msg(const char *msg)
    ++static void write_err(const char *msg)
     +{
    ++	write_in_full(2, "error: ", strlen("error: "));
     +	write_in_full(2, msg, strlen(msg));
     +	write_in_full(2, "\n", 1);
     +}
    @@ compat/terminal.c: static void restore_term_on_signal(int sig)
     +	struct sigaction sa = { .sa_handler = SIG_DFL };
     +
     +	ttou_received = 1;
    -+	write_msg(background_resume_msg);
    ++	write_err(background_resume_msg);
     +	sigaction(signo, &sa, &old_sa);
     +	raise(signo);
     +	sigemptyset(&mask);
    @@ compat/terminal.c: static void restore_term_on_signal(int sig)
     +		can_restore = 0;
     +
     +	if (tcsetattr(term_fd, TCSAFLUSH, &old_term) < 0)
    -+		write_msg(restore_error_msg);
    ++		write_err(restore_error_msg);
     +
     +	sigaction(signo, &sa, &old_sa);
     +	raise(signo);
    @@ compat/terminal.c: static void restore_term_on_signal(int sig)
     +	sigprocmask(SIG_BLOCK, &mask, NULL);
     +	sigaction(signo, &old_sa, NULL);
     +	if (!can_restore) {
    -+		write_msg(restore_error_msg);
    ++		write_err(restore_error_msg);
     +		goto out;
     +	}
     +	/*
    @@ compat/terminal.c: static void restore_term_on_signal(int sig)
     +	if (ttou_received)
     +		goto again;
     +	else if (res < 0)
    -+		write_msg(restore_error_msg);
    ++		write_err(restore_error_msg);
     +	sigaction(SIGTTOU, &old_sa, NULL);
     + out:
     +	errno = saved_errno;
    @@ compat/terminal.c: void restore_term(void)
     +		signal(SIGTTIN, SIG_DFL);
     +		signal(SIGTTOU, SIG_DFL);
     +		signal(SIGTSTP, SIG_DFL);
    -+		FREE_AND_NULL(restore_error_msg);
    -+		FREE_AND_NULL(background_resume_msg);
    ++		restore_error_msg = NULL;
    ++		background_resume_msg = NULL;
     +	}
      }
      
    - int save_term(unsigned flags)
    + int save_term(enum save_term_flags flags)
      {
     +	struct sigaction sa;
     +
      	if (term_fd < 0)
      		term_fd = (flags & SAVE_TERM_STDIN) ? 0
      						    : open("/dev/tty", O_RDWR);
    -@@ compat/terminal.c: int save_term(unsigned flags)
    +@@ compat/terminal.c: int save_term(enum save_term_flags flags)
      	if (tcgetattr(term_fd, &old_term) < 0)
      		return -1;
      	sigchain_push_common(restore_term_on_signal);
    @@ compat/terminal.c: int save_term(unsigned flags)
     +		return 0;
     +
     +	/* avoid calling gettext() from signal handler */
    -+	background_resume_msg = xstrdup(_("error: cannot resume in the background"));
    -+	restore_error_msg = xstrdup(_("error: cannot restore terminal settings"));
    ++	background_resume_msg = _("cannot resume in the background, please use 'fg' to resume");
    ++	restore_error_msg = _("cannot restore terminal settings");
     +	sa.sa_handler = restore_terminal_on_suspend;
     +	sa.sa_flags = SA_RESTART;
     +	sigemptyset(&sa.sa_mask);

Comments

Junio C Hamano March 9, 2022, 11:10 p.m. UTC | #1
Phillip Wood <phillip.wood123@gmail.com> writes:

> V1 Cover Letter:
>
> Fix the remaining issues that I'm aware of when using the built in
> "add -p" with interactive.singlekey that are stopping it from being
> merged to master. The first three patches make sure that we call
> tcsetattr() and the same file descriptor that we use for read() and
> work around poll() being broken when reading from terminals on
> macos. The final patch is more of an improvement rather than a bug fix
> (the same issue already exists in the perl version) and could proceed
> separately.
>
> Unfortunately these patches conflict with
> 'cb/save-term-across-editor-invocation' as well as the textual
> conflicts there is a semantic conflict as the argument to save_term()
> is changed so the code in editor.c will need updating.
>
> These patches are based on 'pw/single-key-interactive'

Is it still true, or does the base only apply to v1?

$ git checkout --detach pw/single-key-interactive
HEAD is now at ac618c418e add -p: disable stdin buffering when interactive.singlekey is set
$ git am -s ./+pw4-v2-add-p-single
Applying: terminal: use flags for save_term()
Applying: terminal: don't assume stdin is /dev/tty
Applying: terminal: work around macos poll() bug
error: patch failed: compat/terminal.c:397
error: compat/terminal.c: patch does not apply
Patch failed at 0003 terminal: work around macos poll() bug
hint: Use 'git am --show-current-patch=diff' to see the failed patch
When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".
$ exit
Junio C Hamano March 9, 2022, 11:37 p.m. UTC | #2
Junio C Hamano <gitster@pobox.com> writes:

>> These patches are based on 'pw/single-key-interactive'
>
> Is it still true, or does the base only apply to v1?
>
> $ git checkout --detach pw/single-key-interactive
> HEAD is now at ac618c418e add -p: disable stdin buffering when interactive.singlekey is set
> $ git am -s ./+pw4-v2-add-p-single
> Applying: terminal: use flags for save_term()
> Applying: terminal: don't assume stdin is /dev/tty
> Applying: terminal: work around macos poll() bug
> error: patch failed: compat/terminal.c:397
> error: compat/terminal.c: patch does not apply
> Patch failed at 0003 terminal: work around macos poll() bug
> hint: Use 'git am --show-current-patch=diff' to see the failed patch
> When you have resolved this problem, run "git am --continue".
> If you prefer to skip this patch, run "git am --skip" instead.
> To restore the original branch and stop patching, run "git am --abort".
> $ exit

I think I figured it out.  A merge of pw/single-key-interactive into
a recent tip of 'master' wants the "return 0" in the preimage below
to be "break" in compat/terminal.c


> @@ -397,12 +433,7 @@ int read_key_without_echo(struct strbuf *buf)
>  		 * half a second when we know that the sequence is complete.
>  		 */
>  		while (!is_known_escape_sequence(buf->buf)) {
> -			struct pollfd pfd = { .fd = 0, .events = POLLIN };
> -
> -			if (poll(&pfd, 1, 500) < 1)
> -				break;
> -
> -			ch = getchar();
> +			ch = getchar_with_timeout(500);
>  			if (ch == EOF)
>  				return 0;
>  			strbuf_addch(buf, ch);

I'll wiggle it in to see how well it fares with other topics in
flight.

Thanks.
Johannes Schindelin March 10, 2022, 1:25 p.m. UTC | #3
Hi Phillip,

On Wed, 9 Mar 2022, Phillip Wood wrote:

> From: Phillip Wood <phillip.wood@dunelm.org.uk>
>
> Thanks to Ramsay and Ævar for their comments on V1.
> Changes since V1:
>  * Patch 1
>    - use an enum for save_term()'s flags (suggested by Ævar)
>    - fixed argument order in the windows code (thanks to Ramsay)
>  * Patch 2
>    - fixed a typo in a comment (thanks to Ramsay)
>  * Patch 4
>    - stopped duplicating the strings returned by gettext() (suggested by
>      Ævar)
>    - reworked error message handling in the signal handler to add an
>      "error: " prefix (suggested by Ævar)
>    - tweaked the background resume error message

While I did not ask for any of these changes, they look good to me. I had
a look over the range-diff and found it reasonable.

Onwards to a bright built-in `add -p` future!

Thanks,
Dscho
Phillip Wood March 10, 2022, 1:28 p.m. UTC | #4
Hi Junio

On 09/03/2022 23:37, Junio C Hamano wrote:
> Junio C Hamano <gitster@pobox.com> writes:
> 
>>> These patches are based on 'pw/single-key-interactive'
>>
>> Is it still true, or does the base only apply to v1?

The base is unchanged but does not seem to match
pw/single-key-interactive. I'm not sure what happened there. They are
based on 300db53b37 ("add -p: disable stdin buffering when
interactive.singlekey is set", 2022-02-16) which is the second parent
of e53fb7aa3f ("Merge branch 'pw/single-key-interactive' into seen",
2022-02-20)

>> $ git checkout --detach pw/single-key-interactive
>> HEAD is now at ac618c418e add -p: disable stdin buffering when interactive.singlekey is set
>> $ git am -s ./+pw4-v2-add-p-single
>> Applying: terminal: use flags for save_term()
>> Applying: terminal: don't assume stdin is /dev/tty
>> Applying: terminal: work around macos poll() bug
>> error: patch failed: compat/terminal.c:397
>> error: compat/terminal.c: patch does not apply
>> Patch failed at 0003 terminal: work around macos poll() bug
>> hint: Use 'git am --show-current-patch=diff' to see the failed patch
>> When you have resolved this problem, run "git am --continue".
>> If you prefer to skip this patch, run "git am --skip" instead.
>> To restore the original branch and stop patching, run "git am --abort".
>> $ exit
> 
> I think I figured it out.  A merge of pw/single-key-interactive into
> a recent tip of 'master' wants the "return 0" in the preimage below
> to be "break" in compat/terminal.c
> 
> 
>> @@ -397,12 +433,7 @@ int read_key_without_echo(struct strbuf *buf)
>>   		 * half a second when we know that the sequence is complete.
>>   		 */
>>   		while (!is_known_escape_sequence(buf->buf)) {
>> -			struct pollfd pfd = { .fd = 0, .events = POLLIN };
>> -
>> -			if (poll(&pfd, 1, 500) < 1)
>> -				break;
>> -
>> -			ch = getchar();
>> +			ch = getchar_with_timeout(500);
>>   			if (ch == EOF)
>>   				return 0;
>>   			strbuf_addch(buf, ch);

That looks good to me. However unfortunately there are some semantic
conflicts as well. The patch below is based on 6b1f77214c ("Merge branch
'pw/add-p-single-key' into seen", 2022-03-09), hopefully Thunderbird wont
mangle it. Whilst preparing the fixup I realized I need to reroll to
fix a closing stdin in patch 2 and resetting the job signals on error in
patch 4. What's the best base to use when rerolling?

Best Wishes

Phillip

---- >8 ----

From: Phillip Wood <phillip.wood@dunelm.org.uk>
Date: Thu, 10 Mar 2022 11:05:26 +0000
Subject: [PATCH] fixup! Merge branch 'pw/add-p-single-key' into seen

---
  compat/terminal.c | 3 ++-
  editor.c          | 2 +-
  2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/compat/terminal.c b/compat/terminal.c
index 3172f4f408..d1ed5c07dc 100644
--- a/compat/terminal.c
+++ b/compat/terminal.c
@@ -141,7 +141,8 @@ int save_term(enum save_term_flags flags)
         if (term_fd < 0)
                 return -1;
         if ((flags & SAVE_TERM_DUPLEX) && !is_controlling_terminal(term_fd)) {
-               close(term_fd);
+               if (term_fd) /* avoid closing stdin */
+                       close(term_fd);
                 term_fd = -1;
                 return -1;
         }
diff --git a/editor.c b/editor.c
index 6c5c95e6a2..192d6ea75d 100644
--- a/editor.c
+++ b/editor.c
@@ -55,7 +55,7 @@ static int prepare_term(const char *editor)
  
         git_config_get_bool("editor.stty", &need_saverestore);
         if (need_saverestore)
-               return save_term(1);
+               return save_term(SAVE_TERM_DUPLEX);
         return 0;
  }
Phillip Wood March 10, 2022, 4:08 p.m. UTC | #5
Hi Dscho

On 10/03/2022 13:25, Johannes Schindelin wrote:
> Hi Phillip,
> 
> On Wed, 9 Mar 2022, Phillip Wood wrote:
> 
>> From: Phillip Wood <phillip.wood@dunelm.org.uk>
>>
>> Thanks to Ramsay and Ævar for their comments on V1.
>> Changes since V1:
>>   * Patch 1
>>     - use an enum for save_term()'s flags (suggested by Ævar)
>>     - fixed argument order in the windows code (thanks to Ramsay)
>>   * Patch 2
>>     - fixed a typo in a comment (thanks to Ramsay)
>>   * Patch 4
>>     - stopped duplicating the strings returned by gettext() (suggested by
>>       Ævar)
>>     - reworked error message handling in the signal handler to add an
>>       "error: " prefix (suggested by Ævar)
>>     - tweaked the background resume error message
> 
> While I did not ask for any of these changes, they look good to me. I had
> a look over the range-diff and found it reasonable.
> 
> Onwards to a bright built-in `add -p` future!

Hopefully! I spotted a couple more small issues earlier so there will be 
a v3 soonish

Best Wishes

Phillip
Phillip Wood March 10, 2022, 6:18 p.m. UTC | #6
On 10/03/2022 13:28, Phillip Wood wrote:
> Hi Junio
> 
> On 09/03/2022 23:37, Junio C Hamano wrote:
>> Junio C Hamano <gitster@pobox.com> writes:
>>
>>> $ git checkout --detach pw/single-key-interactive
>>> HEAD is now at ac618c418e add -p: disable stdin buffering when 
>>> interactive.singlekey is set
>>> $ git am -s ./+pw4-v2-add-p-single
>>> Applying: terminal: use flags for save_term()
>>> Applying: terminal: don't assume stdin is /dev/tty
>>> Applying: terminal: work around macos poll() bug
>>> error: patch failed: compat/terminal.c:397
>>> error: compat/terminal.c: patch does not apply
>>> Patch failed at 0003 terminal: work around macos poll() bug
>>> hint: Use 'git am --show-current-patch=diff' to see the failed patch
>>> When you have resolved this problem, run "git am --continue".
>>> If you prefer to skip this patch, run "git am --skip" instead.
>>> To restore the original branch and stop patching, run "git am --abort".
>>> $ exit
>>
>> I think I figured it out.  A merge of pw/single-key-interactive into
>> a recent tip of 'master' wants the "return 0" in the preimage below
>> to be "break" in compat/terminal.c
>>
>>
>>> @@ -397,12 +433,7 @@ int read_key_without_echo(struct strbuf *buf)
>>>            * half a second when we know that the sequence is complete.
>>>            */
>>>           while (!is_known_escape_sequence(buf->buf)) {
>>> -            struct pollfd pfd = { .fd = 0, .events = POLLIN };
>>> -
>>> -            if (poll(&pfd, 1, 500) < 1)
>>> -                break;
>>> -
>>> -            ch = getchar();
>>> +            ch = getchar_with_timeout(500);
>>>               if (ch == EOF)
>>>                   return 0;

Looking more closely that should be "break" not "return 0". I think what 
has happened is that I accidentally based these on an old version of 
pw/single-key-interactive which did not contain 24d7ce383a ("terminal: 
always reset terminal when reading without echo", 2022-02-22)

Sorry for the confusion

Phillip
Junio C Hamano March 10, 2022, 6:53 p.m. UTC | #7
Phillip Wood <phillip.wood123@gmail.com> writes:

> Looking more closely that should be "break" not "return 0". I think
> what has happened is that I accidentally based these on an old version
> of pw/single-key-interactive which did not contain 24d7ce383a
> ("terminal: always reset terminal when reading without echo",
> 2022-02-22)

I thought the base pw/single-key-interactive topic is solid enough
and ready for 'next'?  It probably is a good idea to rebase these
follow-on patches on top of a merge of the base topic into 'master'.

I'll keep v2 in 'seen' but mark the topic as "Expecting a reroll"
in the draft of the next issue of the "What's cooking" report.

Thanks.