[PATCH 1/2] [gdb] Rename variable main_thread to main_thread_id

[PATCH 1/2] [gdb] Rename variable main_thread to main_thread_id

[Tom de Vries]

I noticed that the variable main_thread:
...
/* The main thread.  */

static std::thread::id main_thread;
...
has a confusing name and corresponding comment, because it doesn't contain the
main thread, but rather the main thread's std::thread::id.

Fix this by renaming to main_thread_id.

Tested on x86_64-linux.
---
 gdb/run-on-main-thread.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/gdb/run-on-main-thread.c b/gdb/run-on-main-thread.c
index c7d9de0afc8..91d25dae28f 100644
--- a/gdb/run-on-main-thread.c
+++ b/gdb/run-on-main-thread.c
@@ -39,9 +39,9 @@ static std::vector<std::function<void ()>> runnables;
 
 static std::mutex runnable_mutex;
 
-/* The main thread.  */
+/* The main thread's thread id.  */
 
-static std::thread::id main_thread;
+static std::thread::id main_thread_id;
 
 #endif
 
@@ -100,7 +100,7 @@ bool
 is_main_thread ()
 {
 #if CXX_STD_THREAD
-  return std::this_thread::get_id () == main_thread;
+  return std::this_thread::get_id () == main_thread_id;
 #else
   return true;
 #endif
@@ -111,7 +111,7 @@ void
 _initialize_run_on_main_thread ()
 {
 #if CXX_STD_THREAD
-  main_thread = std::this_thread::get_id ();
+  main_thread_id = std::this_thread::get_id ();
 #endif
   runnable_event = make_serial_event ();
   add_file_handler (serial_event_fd (runnable_event), run_events, nullptr,

base-commit: 29c108c9610640439daa5244a573348b7c47d994
-- 
2.35.3

[PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom de Vries]

I wrote a patch using is_main_thread (), and found it returning false in the
main thread due to main_thread_id not being initialized yet.

Initialization currently takes place in _initialize_run_on_main_thread, but
that's too late for earlier uses.

Fix this by also initializing on first use.

Tested on x86_64-linux.
---
 gdb/run-on-main-thread.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/gdb/run-on-main-thread.c b/gdb/run-on-main-thread.c
index 91d25dae28f..bee5885d9a6 100644
--- a/gdb/run-on-main-thread.c
+++ b/gdb/run-on-main-thread.c
@@ -94,12 +94,27 @@ run_on_main_thread (std::function<void ()> &&func)
   serial_event_set (runnable_event);
 }
 
+#if CXX_STD_THREAD
+static void
+initialize_main_thread_id (void)
+{
+  static bool initialized = false;
+
+  if (initialized)
+    return;
+  initialized = true;
+
+  main_thread_id = std::this_thread::get_id ();
+}
+#endif
+
 /* See run-on-main-thread.h.  */
 
 bool
 is_main_thread ()
 {
 #if CXX_STD_THREAD
+  initialize_main_thread_id ();
   return std::this_thread::get_id () == main_thread_id;
 #else
   return true;
@@ -111,7 +126,7 @@ void
 _initialize_run_on_main_thread ()
 {
 #if CXX_STD_THREAD
-  main_thread_id = std::this_thread::get_id ();
+  initialize_main_thread_id ();
 #endif
   runnable_event = make_serial_event ();
   add_file_handler (serial_event_fd (runnable_event), run_events, nullptr,
-- 
2.35.3

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom Tromey]

>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:

Tom> I wrote a patch using is_main_thread (), and found it returning false in the
Tom> main thread due to main_thread_id not being initialized yet.

Tom> Initialization currently takes place in _initialize_run_on_main_thread, but
Tom> that's too late for earlier uses.

Tom> Fix this by also initializing on first use.

I wonder if it's enough to have:

static std::thread::id main_thread_id = std::this_thread::get_id ();

Tom> +#if CXX_STD_THREAD
Tom> +static void
Tom> +initialize_main_thread_id (void)

"(void)" is only needed for C, you can just write "()" here.

Tom

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom de Vries]

On 7/28/23 15:10, Tom Tromey wrote:
>>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> Tom> I wrote a patch using is_main_thread (), and found it returning false in the
> Tom> main thread due to main_thread_id not being initialized yet.
> 
> Tom> Initialization currently takes place in _initialize_run_on_main_thread, but
> Tom> that's too late for earlier uses.
> 
> Tom> Fix this by also initializing on first use.
> 
> I wonder if it's enough to have:
> 
> static std::thread::id main_thread_id = std::this_thread::get_id ();
> 

In principle it's enough but it does expose us to the "Static 
Initialization Order Fiasco" [1], so I prefer this solution.

> Tom> +#if CXX_STD_THREAD
> Tom> +static void
> Tom> +initialize_main_thread_id (void)
> 
> "(void)" is only needed for C, you can just write "()" here.
> 

Ack, I'll update it in my reply to Pedro.

Thanks,
- Tom

[1] https://en.cppreference.com/w/cpp/language/siof

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom Tromey]

>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:

>> I wonder if it's enough to have:
>> static std::thread::id main_thread_id = std::this_thread::get_id ();

Tom> In principle it's enough but it does expose us to the "Static
Tom> Initialization Order Fiasco" [1], so I prefer this solution.

If we think it's safe enough to initialize it on first use, just moving
the definition into is_main_thread is enough to avoid this problem.

Now, in theory it is possible for gdb to start a thread very early and
then have that thread call is_main_thread, setting the global
incorrectly.  If we care about that then we need an explicit call to set
it early during startup.

Tom

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom de Vries]

[-- Attachment #1: Type: text/plain, Size: 888 bytes --]

On 8/2/23 18:55, Tom Tromey wrote:
>>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
>>> I wonder if it's enough to have:
>>> static std::thread::id main_thread_id = std::this_thread::get_id ();
> 
> Tom> In principle it's enough but it does expose us to the "Static
> Tom> Initialization Order Fiasco" [1], so I prefer this solution.
> 
> If we think it's safe enough to initialize it on first use, just moving
> the definition into is_main_thread is enough to avoid this problem.
> 

Done.

> Now, in theory it is possible for gdb to start a thread very early and
> then have that thread call is_main_thread, setting the global
> incorrectly.  If we care about that then we need an explicit call to set
> it early during startup.

Yes, Pedro had the same remark, that was already fixed in the previous 
version.

Any further comments?

Thanks,
- Tom


[-- Attachment #2: 0001-gdb-Initialize-main_thread_id-earlier.patch --]
[-- Type: text/x-patch, Size: 2694 bytes --]

From dc19a86c9fd189bb2e7394199e5fda0179d8e6cb Mon Sep 17 00:00:00 2001
From: Tom de Vries <tdevries@suse.de>
Date: Fri, 28 Jul 2023 11:41:07 +0200
Subject: [PATCH] [gdb] Initialize main_thread_id earlier

I wrote a patch using is_main_thread (), and found it returning false in the
main thread due to main_thread_id not being initialized yet.

Initialization currently takes place in _initialize_run_on_main_thread, but
that's too late for earlier uses.

Fix this by initializing, either:
- when entering main, or
- on an earlier first use.

Tested on x86_64-linux.
---
 gdb/gdb.c                |  7 +++++++
 gdb/run-on-main-thread.c | 19 ++++++++++++++++++-
 2 files changed, 25 insertions(+), 1 deletion(-)

diff --git a/gdb/gdb.c b/gdb/gdb.c
index 71a3fd12ba1..6e3ff0755ab 100644
--- a/gdb/gdb.c
+++ b/gdb/gdb.c
@@ -19,10 +19,17 @@
 #include "defs.h"
 #include "main.h"
 #include "interps.h"
+#include "run-on-main-thread.h"
 
 int
 main (int argc, char **argv)
 {
+  /* The first call to is_main_thread () should be from the main thread.
+     If this is the first call, then that requirement is fulfilled here.
+     If this is not the first call, then this verifies that the first call
+     fulfilled that requirement.  */
+  gdb_assert (is_main_thread ());
+
   struct captured_main_args args;
 
   memset (&args, 0, sizeof args);
diff --git a/gdb/run-on-main-thread.c b/gdb/run-on-main-thread.c
index 91d25dae28f..e5c9a775b13 100644
--- a/gdb/run-on-main-thread.c
+++ b/gdb/run-on-main-thread.c
@@ -94,12 +94,24 @@ run_on_main_thread (std::function<void ()> &&func)
   serial_event_set (runnable_event);
 }
 
+#if CXX_STD_THREAD
+static bool main_thread_id_initialized = false;
+#endif
+
 /* See run-on-main-thread.h.  */
 
 bool
 is_main_thread ()
 {
 #if CXX_STD_THREAD
+  /* Initialize main_thread_id on first use of is_main_thread.  */
+  if (!main_thread_id_initialized)
+    {
+      main_thread_id_initialized = true;
+
+      main_thread_id = std::this_thread::get_id ();
+    }
+
   return std::this_thread::get_id () == main_thread_id;
 #else
   return true;
@@ -111,7 +123,12 @@ void
 _initialize_run_on_main_thread ()
 {
 #if CXX_STD_THREAD
-  main_thread_id = std::this_thread::get_id ();
+  /* The variable main_thread_id should be initialized when entering main, or
+     at an earlier use, so it should already be initialized here.  */
+  gdb_assert (main_thread_id_initialized);
+
+  /* Assume that we execute this in the main thread.  */
+  gdb_assert (is_main_thread ());
 #endif
   runnable_event = make_serial_event ();
   add_file_handler (serial_event_fd (runnable_event), run_events, nullptr,

base-commit: 86bb2d76c901a558e5b2dd29df95874ce6e78dae
-- 
2.35.3

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom Tromey]

>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:

>> If we think it's safe enough to initialize it on first use, just
>> moving
>> the definition into is_main_thread is enough to avoid this problem.
>> 

Tom> Done.

This wasn't really what I meant but it's not important.

Approved-By: Tom Tromey <tom@tromey.com>

Tom

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom de Vries]

On 8/2/23 22:48, Tom Tromey wrote:
>>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
>>> If we think it's safe enough to initialize it on first use, just
>>> moving
>>> the definition into is_main_thread is enough to avoid this problem.
>>>
> 
> Tom> Done.
> 
> This wasn't really what I meant but it's not important.
> 

Thanks for the review.

Sorry for misunderstanding this, if you'd like me to follow up in any 
way, let me know.

Committed.

Thanks,
- Tom

> Approved-By: Tom Tromey <tom@tromey.com>
> 
> Tom

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Pedro Alves]

On 2023-07-28 11:59, Tom de Vries via Gdb-patches wrote:
> I wrote a patch using is_main_thread (), and found it returning false in the
> main thread due to main_thread_id not being initialized yet.
> 
> Initialization currently takes place in _initialize_run_on_main_thread, but
> that's too late for earlier uses.
> 
> Fix this by also initializing on first use.
> 
> Tested on x86_64-linux.
> ---
>  gdb/run-on-main-thread.c | 17 ++++++++++++++++-
>  1 file changed, 16 insertions(+), 1 deletion(-)
> 
> diff --git a/gdb/run-on-main-thread.c b/gdb/run-on-main-thread.c
> index 91d25dae28f..bee5885d9a6 100644
> --- a/gdb/run-on-main-thread.c
> +++ b/gdb/run-on-main-thread.c
> @@ -94,12 +94,27 @@ run_on_main_thread (std::function<void ()> &&func)
>    serial_event_set (runnable_event);
>  }
>  
> +#if CXX_STD_THREAD
> +static void
> +initialize_main_thread_id (void)
> +{
> +  static bool initialized = false;
> +
> +  if (initialized)
> +    return;
> +  initialized = true;
> +
> +  main_thread_id = std::this_thread::get_id ();
> +}
> +#endif
> +
>  /* See run-on-main-thread.h.  */
>  
>  bool
>  is_main_thread ()
>  {
>  #if CXX_STD_THREAD
> +  initialize_main_thread_id ();

This is assuming that is_main_thread() will always be called once by
the main thread, before any other thread could call it.  Otherwise if
is_main_thread is called by some other thread first,
is_main_thread -> initialize_main_thread_id will record the wrong thread
id in the main_thread_id.  At least a comment somewhere would
be warranted.  I think putting:

  gdb_assert (is_main_thread ());

somewhere in the early main code would be good.  It would prevent such
situation from ever happening undetected.

Re: [PATCH 2/2] [gdb] Initialize main_thread_id earlier

[Tom de Vries]

[-- Attachment #1: Type: text/plain, Size: 1908 bytes --]

On 7/28/23 17:36, Pedro Alves wrote:
> On 2023-07-28 11:59, Tom de Vries via Gdb-patches wrote:
>> I wrote a patch using is_main_thread (), and found it returning false in the
>> main thread due to main_thread_id not being initialized yet.
>>
>> Initialization currently takes place in _initialize_run_on_main_thread, but
>> that's too late for earlier uses.
>>
>> Fix this by also initializing on first use.
>>
>> Tested on x86_64-linux.
>> ---
>>   gdb/run-on-main-thread.c | 17 ++++++++++++++++-
>>   1 file changed, 16 insertions(+), 1 deletion(-)
>>
>> diff --git a/gdb/run-on-main-thread.c b/gdb/run-on-main-thread.c
>> index 91d25dae28f..bee5885d9a6 100644
>> --- a/gdb/run-on-main-thread.c
>> +++ b/gdb/run-on-main-thread.c
>> @@ -94,12 +94,27 @@ run_on_main_thread (std::function<void ()> &&func)
>>     serial_event_set (runnable_event);
>>   }
>>   
>> +#if CXX_STD_THREAD
>> +static void
>> +initialize_main_thread_id (void)
>> +{
>> +  static bool initialized = false;
>> +
>> +  if (initialized)
>> +    return;
>> +  initialized = true;
>> +
>> +  main_thread_id = std::this_thread::get_id ();
>> +}
>> +#endif
>> +
>>   /* See run-on-main-thread.h.  */
>>   
>>   bool
>>   is_main_thread ()
>>   {
>>   #if CXX_STD_THREAD
>> +  initialize_main_thread_id ();
> 

Hi Pedro,

thanks for the review.

> This is assuming that is_main_thread() will always be called once by
> the main thread, before any other thread could call it.  Otherwise if
> is_main_thread is called by some other thread first,
> is_main_thread -> initialize_main_thread_id will record the wrong thread
> id in the main_thread_id.  At least a comment somewhere would
> be warranted.  I think putting:
> 
>    gdb_assert (is_main_thread ());
> 
> somewhere in the early main code would be good.  It would prevent such
> situation from ever happening undetected.
> 

Updated accordingly.  Any further comments?

Thanks,
- Tom

[-- Attachment #2: 0001-gdb-Initialize-main_thread_id-earlier.patch --]
[-- Type: text/x-patch, Size: 2775 bytes --]

From 3886892551c9e35b6ea36a6df23167fa3690a319 Mon Sep 17 00:00:00 2001
From: Tom de Vries <tdevries@suse.de>
Date: Fri, 28 Jul 2023 11:41:07 +0200
Subject: [PATCH] [gdb] Initialize main_thread_id earlier

I wrote a patch using is_main_thread (), and found it returning false in the
main thread due to main_thread_id not being initialized yet.

Initialization currently takes place in _initialize_run_on_main_thread, but
that's too late for earlier uses.

Fix this by initializing, either:
- when entering main, or
- on an earlier first use.

Tested on x86_64-linux.
---
 gdb/gdb.c                |  7 +++++++
 gdb/run-on-main-thread.c | 24 +++++++++++++++++++++++-
 2 files changed, 30 insertions(+), 1 deletion(-)

diff --git a/gdb/gdb.c b/gdb/gdb.c
index 71a3fd12ba1..6e3ff0755ab 100644
--- a/gdb/gdb.c
+++ b/gdb/gdb.c
@@ -19,10 +19,17 @@
 #include "defs.h"
 #include "main.h"
 #include "interps.h"
+#include "run-on-main-thread.h"
 
 int
 main (int argc, char **argv)
 {
+  /* The first call to is_main_thread () should be from the main thread.
+     If this is the first call, then that requirement is fulfilled here.
+     If this is not the first call, then this verifies that the first call
+     fulfilled that requirement.  */
+  gdb_assert (is_main_thread ());
+
   struct captured_main_args args;
 
   memset (&args, 0, sizeof args);
diff --git a/gdb/run-on-main-thread.c b/gdb/run-on-main-thread.c
index 91d25dae28f..5d453ffe612 100644
--- a/gdb/run-on-main-thread.c
+++ b/gdb/run-on-main-thread.c
@@ -94,12 +94,29 @@ run_on_main_thread (std::function<void ()> &&func)
   serial_event_set (runnable_event);
 }
 
+#if CXX_STD_THREAD
+static bool main_thread_id_initialized = false;
+
+static void
+initialize_main_thread_id ()
+{
+  if ( main_thread_id_initialized)
+    return;
+   main_thread_id_initialized = true;
+
+  main_thread_id = std::this_thread::get_id ();
+}
+#endif
+
 /* See run-on-main-thread.h.  */
 
 bool
 is_main_thread ()
 {
 #if CXX_STD_THREAD
+  /* Initialize main_thread_id on first use of is_main_thread.  */
+  initialize_main_thread_id ();
+
   return std::this_thread::get_id () == main_thread_id;
 #else
   return true;
@@ -111,7 +128,12 @@ void
 _initialize_run_on_main_thread ()
 {
 #if CXX_STD_THREAD
-  main_thread_id = std::this_thread::get_id ();
+  /* The variable main_thread_id should be initialized when entering main, or
+     at an earlier use, so it should already be initialized here.  */
+  gdb_assert (main_thread_id_initialized);
+
+  /* Assume that we execute this in the main thread.  */
+  gdb_assert (is_main_thread ());
 #endif
   runnable_event = make_serial_event ();
   add_file_handler (serial_event_fd (runnable_event), run_events, nullptr,

base-commit: b5c37946cce4b41af0436529b139fa3d23e61f73
-- 
2.35.3

Re: [PATCH 1/2] [gdb] Rename variable main_thread to main_thread_id

[Tom Tromey]

>>>>> "Tom" == Tom de Vries via Gdb-patches <gdb-patches@sourceware.org> writes:

Tom> I noticed that the variable main_thread:
Tom> ...
Tom> /* The main thread.  */

Tom> static std::thread::id main_thread;
Tom> ...
Tom> has a confusing name and corresponding comment, because it doesn't contain the
Tom> main thread, but rather the main thread's std::thread::id.

Tom> Fix this by renaming to main_thread_id.

Thanks.
Approved-By: Tom Tromey <tom@tromey.com>

Tom