async with message.channel.typing() only shows indicator for 10 seconds

[Octopirate1]

Summary

In the docs, async with message.channel.typing() says that the typing indicator is indefinitely shown until the with context is exited. In reality, it is either 10 seconds or when the with context is exited, whichever comes first.

Reproduction Steps

    if message.content.startswith('!type'):
        async with message.channel.typing():
            time.sleep(20) # or asyncio
            await message.channel.send("Time up")

Here, there should be a 20 second delay, during which the typing indicator in discord is active, and then a message saying "Time up". Instead, the indicator deactivates halfway through the sleep.

Checklist

• I have searched the open issues for duplicates.
• [N/A] I have shown the entire traceback, if possible.
• I have removed my token from display, if visible.
• I have provided the intents that my bot is using.

System Information

• Python v3.5.3-final
• discord.py v1.5.1-final
• aiohttp v3.6.3
• system info: Linux 4.4.0-19041-Microsoft Fix issue with author changing type #1-Microsoft Fri Dec 06 14:06:00 PST 2019

[Rapptz]

Please read the FAQ on blocking.

By using time.sleep you disallow the bot from doing anything, this includes sending the second request to continue the typing indicator.

[Octopirate1]

This is a snippet of my real code:

async with message.channel.typing():

            messages = await message.channel.history(limit=messagecontext).flatten()

            messagearray = ["[{}] {}#{}\n{}\n\n\n".format(message.created_at.strftime("%d-%b-%y %I:%M %p"), message.author.name, message.author.discriminator, message.content) for message in messages[1:][::-1]]
            messagearray.append("[{}] {}#{}\n".format(message.created_at.strftime("%d-%b-%y %I:%M %p"), message.author.name, message.author.discriminator))

            currenttime = "[{}] ".format(message.created_at.strftime("%d-%b-%y %I:%M %p"))

            print(messagearray)

            if "response" in seccommand or "respond" in seccommand:
                responsearray = generate_resp(''.join(messagearray)) # where the majority of time is spent

                responsesplits = [[responsesample.split("\n\n\n")] for responsesample in responsearray]
                responses = [response[0][messagecontext - 1] for response in responsesplits] # response[0] because of extra dimension
                global cutoff
                cutoff = True if responsesplits[0][0][messagecontext - 1] == responsesplits[0][0][-1] and responsesplits[0][0][messagecontext - 1][-1] != "\n" else False
                await message.channel.send(responses[0])

where

def generate_resp(text):
    response = gpt2.generate(sess,
              length=messagecontext*10,
              temperature=temp,
              prefix=text,
              return_as_list=True,
              # nsamples=2
              # batch_size=3
              )
    return response

generate_resp is a gpt-2 instance that is generating text. I've tried using async def generate_resp(text) and awaiting on it but that didn't seem to solve the issue at all. Should I be trying another asynchronous approach?

[Rapptz]

Since this is a blocking call then you need to turn it into an asynchronous one, presumably using run_in_executor on the loop.

For example:

responsearray = client.loop.run_in_executor(None, generate_resp, ''.join(messagearray)) # where the majority of time is spent

Where client is replaced with the name of whatever you called your discord.Client or commands.Bot instance.

[Octopirate1]

While I believe this would work if it was any other function, TensorFlow comes with it's own problems (as it frequently does) with run_in_executor - that, however, I will solve myself. Thank you for your time.