Always write io_buffer when in "enum bodies" branch.

[collinsauve]

Description

Root of the issue is that you can get into the if chunked branch when body.each has nothing in it, so the previously populated io_buffer is never written to the socket.

This used to be the case but was changed on #3072. I'm just moving this line back.

Closes #3112.

Your checklist for this pull request

• I have reviewed the guidelines for contributing to this repository.
• I have added (or updated) appropriate tests if this PR fixes a bug or adds a feature.
• My pull request is 100 lines added/removed or less so that it can be easily reviewed.
• If this PR doesn't need tests (docs change), I added [ci skip] to the title of the PR.
• If this closes any issues, I have added "Closes #issue" to the PR description or my commit messages.
• I have updated the documentation accordingly.
• All new and existing tests passed, including Rubocop.

[dmke]

I have applied this patch locally and can confirm that head(404) and render(plain: "", status: 404) now behave as expected.

I don't have access to the codebase where I've noticed this issue, so I can't tell if that broke something else 👀

[MSP-Greg]

@collinsauve

Thank you for the PR. I added three tests in a commit at 1aa498d, feel free to use them.

The two array body tests pass, the enum body test fails as shown below:

  1) Failure:
TestPumaServer#test_empty_body_enum [/mnt/c/Greg/GitHub/puma/test/test_puma_server.rb:1567]:
--- expected
+++ actual
@@ -1,6 +1,5 @@
-"HTTP/1.1 404 Not Found\r
-Transfer-Encoding: chunked\r
-\r
-0\r
+# encoding: ASCII-8BIT
+#    valid: true
+"0\r
 \r
"

I also added commit 91ad03d, which has some possible changes for request.rb. When sending requests, Puma doesn't want to buffer a large amount of string data before writing to the client socket, but there is also an interest in minimizing the number of writes. This is especially true with SSL connections.

The code in the commit acts the same as your code when the body is empty, but will combine the header section with the first segment of the body if it isn't empty. There are two other changes that check for nil members of the body.

Thanks again, Greg

EDIT: changed the commits, as Windows Rubies 3.1 and later failed...

[collinsauve]

Thank you @MSP-Greg! I've cherry-picked those commits onto this branch

[MSP-Greg]

@collinsauve

Thank you for all your work.