[rowillia]

This should fix #338

Example:

import asyncio

import httpx

async def await_url(c):
    r = await c.get('http://httpbin.org/get')
    await r.read()
    print(r.elapsed)

async def httpx_test():
    client = httpx.AsyncClient()
    await asyncio.gather(*[await_url(client) for _ in range(25)])

asyncio.run(httpx_test())

[florimondmanca]

Thanks for tackling this @rowillia! I’ll give this a review later. :)

(Edited) One question I had — did you investigate if it was possible to call the elapsed time tracker at a higher level, eg in BaseClient._get_response()? The response object in there is obtained when the headers have been received, which seems to be what we want to be timing for. Of course we'd only get an approximation because elapsed would also include our own processing time, but that's probably negligible anyway? Then we can just set response.elapsed instead of passing it as an argument. The main motivation is to centralize the timing logic in one place, making it less bug-prone and more easy to extend/maintain.

[rowillia]

Thanks for the feedback @florimondmanca ! I went ahead and moved that logic into _get_response.

[sethmlarson]

Btw when we document this let's put a warning to use timedelta.total_seconds() instead of .seconds. This is a common issue that new users of the datetime module hit.

[florimondmanca]

This is great stuff! I left a couple of polishing suggestions. :)

[florimondmanca]

Awesome 🎉 Just a tiny change you can commit via the suggestion, and then we're good to merge I think!

[rowillia]

Thanks for the feedback @florimondmanca ! LMK if there's anything else to update.

[florimondmanca]

I think we're good — merged! Thank you @rowillia. 💯