gh-42550: Add 'Expect: 100-Continue' support to httplib by IngridMorstrad · Pull Request #133276 · python/cpython

IngridMorstrad · 2025-05-01T22:26:29Z

Previously, http.client would always send content body immediately and ignore any 100 responses. This change makes HTTPClient.request() wait for a Continue response if the Expect: 100-Continue header is set, and adds a parameter to HTTPClient.getresponse() that will cause it to return 100 responses instead of eating them.

#42550
https://bugs.python.org/issue1346874

📚 Documentation preview 📚: https://cpython-previews--133276.org.readthedocs.build/

Issue: httplib simply ignores CONTINUE #42550

Previously, http.client would always send content body immediately and ignore any 100 responses. This change makes HTTPClient.request() wait for a `Continue` response if the `Expect: 100-Continue` header is set, and adds a parameter to HTTPClient.getresponse() that will cause it to return 100 responses instead of eating them.

ZeroIntensity

General triage nits.

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>

IngridMorstrad · 2025-08-13T22:55:30Z

Hi, checking for a possible review on this. @ZeroIntensity

ZeroIntensity · 2025-08-14T14:26:39Z

Sorry, I was only triaging. I don't have enough knowledge on the http module to review this. You could take a look at the experts index.

AA-Turner · 2025-08-14T18:50:51Z

+   sending a file-like message body. Finally, the optional *continue_timeout*
+   parameter controls how long the connection will wait for a ``100 Continue``
+   response from the server before sending the body, if the request included
+   an ``Expect: 100-Continue`` header.


Does this need to be a constructor parameter?

I was thinking that since it is a property of the server (how quickly it responds with 100 Continue, or whether it understands Expect at all), it should be in the constructor, not the individual request. The other per-request parameters (expect_continue, ignore_100_continue) could vary per-request. LMK if that makes sense or we need to change this and I can update the PR.

AA-Turner · 2025-08-14T18:51:40Z

You have merge conflicts, please resolve them before requesting a re-review.

IngridMorstrad · 2026-02-11T17:53:48Z

Sorry for the delay here, I just replied to the question above. If that response makes sense, I can send an updated PR with the merge conflicts addressed.

IngridMorstrad · 2026-03-02T22:05:27Z

Hi,

I have fixed the merge conflicts and CI is green. The PR has been open since May 2025. I am not sure who to request a review from as there is no reviewer listed for the http module on the Developer's Guide. Any help is appreciated, thanks!

Ashwin

IngridMorstrad · 2026-03-20T10:53:47Z

Hi @orsenthil , could I please get a review of this? CI is green and I keep needing to rebase over time.

IngridMorstrad · 2026-03-20T10:56:44Z

CC @gpshead as the networking head listed here, wondering if you could help me get a reviewer for this PR.

github-actions · 2026-04-24T06:38:09Z

This PR is stale because it has been open for 30 days with no activity.

gpshead

Left a pile of comments.

We've gone forever without this in our HTTP client. It should be possible to make work, but it might get weird in our stack in the modern TLS HTTPSConnection that everyone should be using instead of HTTPConnection... So maybe something to figure out there. Does this actually work properly in that scenario?

gpshead · 2026-05-18T18:44:02Z

        return version, status, reason

-    def begin(self, *, _max_headers=None):
+    def begin(self, ignore_100_continue=True, _max_headers=None):


the * was important, it makes these keyword only arguments which are easier to support on public APIs like this method. (rule of thumb, when adding a new parameter it either has to be added at the end and if there are either a lot of arguments or are very infrequently passed should be added as keyword only so that it must be passed by name)

Suggested change

def begin(self, ignore_100_continue=True, _max_headers=None):

def begin(self, *, ignore_100_continue=True, _max_headers=None):

gpshead · 2026-05-18T18:46:19Z


    def __init__(self, host, port=None, timeout=socket._GLOBAL_DEFAULT_TIMEOUT,
-                 source_address=None, blocksize=8192, *, max_response_headers=None):
+                 source_address=None, blocksize=8192, continue_timeout=2.5, *,


continue_timeout should be a keyword only argument.... but is the constructur really where we want it rather than on the request() method where headers are specified? That'd also avoid the need to plumb this through HTTPSConnection.__init__.

From an API design perspective, on the HTTPConnection.request method which takes a headers={} dict as well as some flags such as potentially this... If someone wants to use this feature, it is ergonomically nicer if we had a flag on the request_api that explicitly adds the correct Expect: 100-Continue header rather than expecting people to fill in that level of implementation detail header value directly themselves. (I think the code still needs detect if the header was added manually regardless, kind of like it does for some of the other headers)

also, assert that continue_timeout > 0.0.

gpshead · 2026-05-18T18:58:07Z

+    def getresponse(self, ignore_100_continue=True):
        """Get the response from the server.

        If the HTTPConnection is in the correct state, returns an
-        instance of HTTPResponse.
+        instance of HTTPResponse or of whatever object is returned by the
+        response_class variable. The connection will wait for a response other
+        than code 100 ('Continue') unless ignore_100_continue is set to False.


"ignore..." is a somewhat of a negative term in English for use in an API which leads to awkward descriptions of the behavior. it defaults to True and people have to turn the negative off with False leading to double negatives to reason about.

As this defaults to not supporting 100 continue behavior, why not call it support_continue=False so that enabling support for it involves setting something to True instead of setting an ignore= to False?

(apply this everywhere)

also we should make it keyword only so that it has to be spelled out when anyone wants to use it for better code readability. (blah.getresponse(True) doesn't indicate much to the reader).

I also think eliding the number from the name is a good idea. 100 is defined as continue so no need to mention that internal http detail.

gpshead · 2026-05-18T18:59:35Z

    def create_connection(self, *pos, **kw):
        return FakeSocket(*self.fake_socket_args)

+    def fake_select(self, rlist, wlist, xlist, timeout=None):


this doesn't use self, make it a @staticmethod or a top level function?

gpshead · 2026-05-18T19:11:32Z

    >>> print(response.status, response.reason)
    200, OK

+Since version 3.15, conditional transmission of the body is supported when an


don't mention the version here. think of docs as being read in the future, version specific changes are normally footnotes via the version added and version changed markers. (this is a feature so it won't be in 3.15)

gpshead · 2026-05-18T20:37:51Z


        if message_body is not None:
+            if expect_continue and not self.__response:
+                read_ready, _, _ = self._select([self.sock], [], [],


select won't work properly on a HTTPSConnection. There's a .pending method on SSLSocket that could be used in this scenario if read_ready.

Investigate what is plausible and add some tests covering the HTTPSConnection test case.

gpshead · 2026-05-18T20:43:36Z

+                    else:
+                        response = self.response_class(self.sock,
+                                                       method=self._method)
+                    response.begin(ignore_100_continue=False)


_max_headers= should also be passed here.

gpshead · 2026-05-18T20:47:05Z

+                    response.begin(ignore_100_continue=False)
+                    if response.code != CONTINUE:
+                        # Break without sending the body
+                        self.__pending_response = response


wherever self.__response is cleared (= None) such as putrequest, self.__pending_response needs to cleared as well. tests should cover that this happens by running through a scenario that'd lead to this such as maybe a non-100 early response that gets ignored before another request is sent on the same connection?

gpshead · 2026-05-18T21:35:03Z

@@ -0,0 +1,2 @@
+:class:`http.client.HTTPConnection` now waits for a ``100 Continue`` response


We should highlight that this is a behavior change for any code that explicitly added the Extra: 100-Continue to their request headers as those will now wait for the timeout to elapse before sending the body if the server doesn't do 100s.

Also add an entry to the whatsnew 3.16.rst doc.

bedevere-app Bot added the awaiting review label May 1, 2025

IngridMorstrad changed the title ~~Add 'Expect: 100-Continue' support to httplib~~ gh-42550: Add 'Expect: 100-Continue' support to httplib May 1, 2025

bedevere-app Bot mentioned this pull request May 1, 2025

httplib simply ignores CONTINUE #42550

Open

IngridMorstrad mentioned this pull request May 1, 2025

bpo-1346874: Add 'Expect: 100-Continue' support to httplib #14880

Open

ZeroIntensity reviewed May 18, 2025

View reviewed changes

IngridMorstrad and others added 2 commits May 21, 2025 15:23

Update version this will go into

532026d

Co-authored-by: Peter Bierma <zintensitydev@gmail.com>

Merge branch 'main' into fix-issue-42550

7146d5b

IngridMorstrad requested a review from ZeroIntensity June 10, 2025 05:58

AA-Turner reviewed Aug 14, 2025

View reviewed changes

Merge branch 'main' into fix-issue-42550

d902673

merwok requested a review from orsenthil March 3, 2026 01:53

github-actions Bot added the stale Stale PR or inactive for long period of time. label Apr 24, 2026

gpshead reviewed May 18, 2026

View reviewed changes

gpshead removed the stale Stale PR or inactive for long period of time. label May 18, 2026

	def begin(self, ignore_100_continue=True, _max_headers=None):
	def begin(self, *, ignore_100_continue=True, _max_headers=None):

		@@ -0,0 +1,2 @@
		:class:`http.client.HTTPConnection` now waits for a ``100 Continue`` response

Uh oh!

Conversation

IngridMorstrad commented May 1, 2025 • edited by bedevere-app Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ZeroIntensity left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

IngridMorstrad commented Aug 13, 2025

Uh oh!

ZeroIntensity commented Aug 14, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

AA-Turner commented Aug 14, 2025

Uh oh!

IngridMorstrad commented Feb 11, 2026

Uh oh!

IngridMorstrad commented Mar 2, 2026

Uh oh!

IngridMorstrad commented Mar 20, 2026

Uh oh!

IngridMorstrad commented Mar 20, 2026

Uh oh!

github-actions Bot commented Apr 24, 2026

Uh oh!

gpshead left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

IngridMorstrad commented May 1, 2025 •

edited by bedevere-app Bot

Loading