Re: [PATCH v5 2/3] qga: Add `merged` variant to GuestExecCaptureOutputMode

["Daniel Xu" <dxu@dxuuu.xyz>]

Hi Daniel,

Sorry about the delay -- was out of town the past week.

On Fri, Mar 10, 2023, at 2:24 AM, Daniel P. Berrangé wrote:
> On Thu, Mar 09, 2023 at 03:40:57PM -0700, Daniel Xu wrote:
>> Currently, any captured output (via `capture-output`) is segregated into
>> separate GuestExecStatus fields (`out-data` and `err-data`). This means
>> that downstream consumers have no way to reassemble the captured data
>> back into the original stream.
>> 
>> This is relevant for chatty and semi-interactive (ie. read only) CLI
>> tools.  Such tools may deliberately interleave stdout and stderr for
>> visual effect. If segregated, the output becomes harder to visually
>> understand.
>> 
>> This commit adds a new enum variant to the GuestExecCaptureOutputMode
>> qapi to merge the output streams such that consumers can have a pristine
>> view of the original command output.
>> 
>> Signed-off-by: Daniel Xu <dxu@dxuuu.xyz>
>> ---
>>  qga/commands.c       | 31 +++++++++++++++++++++++++++++--
>>  qga/qapi-schema.json |  4 +++-
>>  2 files changed, 32 insertions(+), 3 deletions(-)
>> 
>> diff --git a/qga/commands.c b/qga/commands.c
>> index 01f68b45ab..c347d434ed 100644
>> --- a/qga/commands.c
>> +++ b/qga/commands.c
>> @@ -270,12 +270,26 @@ static void guest_exec_child_watch(GPid pid, gint status, gpointer data)
>>      g_spawn_close_pid(pid);
>>  }
>>  
>> -/** Reset ignored signals back to default. */
>>  static void guest_exec_task_setup(gpointer data)
>>  {
>>  #if !defined(G_OS_WIN32)
>> +    bool has_merge = *(bool *)data;
>>      struct sigaction sigact;
>>  
>> +    if (has_merge) {
>> +        /*
>> +         * FIXME: When `GLIB_VERSION_MIN_REQUIRED` is bumped to 2.58+, use
>> +         * g_spawn_async_with_fds() to be portable on windows. The current
>> +         * logic does not work on windows b/c `GSpawnChildSetupFunc` is run
>> +         * inside the parent, not the child.
>> +         */
>> +        if (dup2(STDOUT_FILENO, STDERR_FILENO) != 0) {
>> +            slog("dup2() failed to merge stderr into stdout: %s",
>> +                 strerror(errno));
>> +        }
>> +    }
>> +
>> +    /* Reset ignored signals back to default. */
>>      memset(&sigact, 0, sizeof(struct sigaction));
>>      sigact.sa_handler = SIG_DFL;
>>  
>> @@ -409,6 +423,7 @@ GuestExec *qmp_guest_exec(const char *path,
>>      GIOChannel *in_ch, *out_ch, *err_ch;
>>      GSpawnFlags flags;
>>      bool has_output = false;
>> +    bool has_merge = false;
>
> Wrap in  #ifndef _WIN32

I think it would be better to leave this variable un-gated b/c gating it
would make the later call to g_spawn_async_with_pipes() less clean.
I don't think it'll trigger any unused variable warnings either since we
are technically always using it.

>
>>      GuestExecCaptureOutputMode output_mode;
>>      g_autofree uint8_t *input = NULL;
>>      size_t ninput = 0;
>> @@ -445,13 +460,25 @@ GuestExec *qmp_guest_exec(const char *path,
>>      case GUEST_EXEC_CAPTURE_OUTPUT_MODE_SEPARATED:
>>          has_output = true;
>>          break;
>> +    case GUEST_EXEC_CAPTURE_OUTPUT_MODE_MERGED:
>> +        has_output = true;
>> +        has_merge = true;
>> +        break;
>
> Wrap in  #ifndef _WIN32
>
>>      case GUEST_EXEC_CAPTURE_OUTPUT_MODE__MAX:
>>          /* Silence warning; impossible branch */
>>          break;
>>      }
>>  
>> +#if defined(G_OS_WIN32)
>> +    /* FIXME: see comment in guest_exec_task_setup() */
>> +    if (has_merge) {
>> +        error_setg(errp, "merged unsupported on windows");
>> +        return NULL;
>> +    }
>> +#endif
>
> THis can be dropped, since 'has_merge' won't exist for
> Win32 builds.
>
>> +
>>      ret = g_spawn_async_with_pipes(NULL, argv, envp, flags,
>> -            guest_exec_task_setup, NULL, &pid, input_data ? &in_fd : NULL,
>> +            guest_exec_task_setup, &has_merge, &pid, input_data ? &in_fd : NULL,
>>              has_output ? &out_fd : NULL, has_output ? &err_fd : NULL, &gerr);
>>      if (!ret) {
>>          error_setg(errp, QERR_QGA_COMMAND_FAILED, gerr->message);
>> diff --git a/qga/qapi-schema.json b/qga/qapi-schema.json
>> index d1e00a4234..b4782525ae 100644
>> --- a/qga/qapi-schema.json
>> +++ b/qga/qapi-schema.json
>> @@ -1210,11 +1210,13 @@
>>  # @stderr: only capture stderr
>>  # @separated: capture both stdout and stderr, but separated into
>>  #             GuestExecStatus out-data and err-data, respectively
>> +# @merged: capture both stdout and stderr, but merge together
>> +#          into out-data. not effective on windows guests.
>>  #
>>  # Since: 8.0
>>  ##
>>   { 'enum': 'GuestExecCaptureOutputMode',
>> -   'data': [ 'none', 'stdout', 'stderr', 'separated' ] }
>> +   'data': [ 'none', 'stdout', 'stderr', 'separated', 'merged' ] }
>
> Actually, I've just realized we can make this conditional:
>
>
>  'data': [ 'none', 'stdout', 'stderr', 'separated',
>            { 'name': 'merged', 'if': 'CONFIG_WIN32' } ] }
>
> so the constant doesn't even exist in Win32 builds.

Ack, that looks cleaner.

[...]

Thanks,
Daniel