replace the heavyhanded connection:close added in b4fddbc3d
with a comparison of content-length to num bytes consumed
this approach also covers incorrectly configured servers
where the reverseproxy was not detected
also adds explicit TE/CL handling, even though most
(all?) reverseproxies already prevent such issues
also adds explicit sanchk of up2k chunk-receiver,
in case any bugs are ever added there
I found it slightly cumbersome to write a config when the only feedback OpenRC gives is `crashed`, so I added a small QOL function to run copyparty with the `--exit cfg` flag directly from the service.
Signed-off-by: charel <charel@lotsaspaghetti.com>
if both `quota-available-bytes` and `quotaused` are ignored
(not even returned as 404), then macos Finder is able to
connect instantly, avoiding this longstanding bug in macos
the presence of `quotaused` is the trigger for this logic, which
is a property apple invented and only apple uses, meaning we can
safely break the webdav spec as required in this case
thx @freddyheppell for the observation
npm is no longer able to build it, and the build output
never changed since copyparty v0.11.20 (2021-06-20) /
asmcrypto 2821dd1dedd1196c378f5854037dda5c869313f3 (2018-12-02)
one final pass of reasonable size-golfing was done by hand
deflated with pigz -11 -I250
Important changes:
The "shift" key is called "MAJ" (short for "majuskla" for "uppercase") in the translation, so the new lines have been updated.
"beligi json" is technically a valid string, but the "-on" ending could be confused for a grammatical ending, with the whole string being interpreted as "beautify javascript" instead, so instead "JSON" is capitalized and another "on" ending is added.
A few commas are added for readability.
Signed-off-by: /dev/urandom <53902042+slashdevslashurandom@users.noreply.github.com>
makes it possible to login from the webbrowser dillo;
* unlike every other browser, dillo does NOT send a trailing "\r\n"
after the terminating "--"; turns out that dillo got this right
and every other browser didn't, fun
* dillo announces the boundary in quotes, which is spec-optional
the multipart-parser is now 2% slower
