I did this

-T path will stat path to figure out its size in bytes to use it as Content-Length if it is a regular file.

The problem with that is that, on BSDs and some other UNIXes (not Linux), open(path) may not give you a file descriptor with a 0 offset from the start of the file.

On BSDs, if the path given to open is to a /dev/fd/* file or /dev/stdin, /dev/stdout, /dev/stderr, that will trigger a special behaviour for open: open on BSD, instead of reopening the file related to that file descriptor with different open modes like in Linux, will actually duplicate the file descriptor specified by that path if its open modes are compatible with the ones requested by open, and otherwise fail.

Since -T /dev/fd/4 for example, will effectively just be a dup(4), when curl will open the path it thinks is for a regular file (because fd 4 is indeed a file descriptor for a regular file, myfile), it will not find itself at byte offset 0 into the file, and will send less data than Content-Length.

printf 'NAME\tJOB\nTom\tCarpenter\nAlice\tMechanic\n' > myfile
{
   read -r skip_firstline <&4 &&
   curl -T /dev/fd/4 localhost:43000
} 4< myfile

$ nc -l localhost:43000
PUT /4 HTTP/1.1
Host: localhost:43000
User-Agent: curl/8.2.1
Accept: */*
Content-Length: 38

Tom     Carpenter
Alice   Mechanic

I expected the following

I expected:

$ nc -l 43000
PUT /4 HTTP/1.1
Host: localhost:43000
User-Agent: curl/8.2.1
Accept: */*
Content-Length: 29

Tom     Carpenter
Alice   Mechanic

curl should checked the current offset into the regular file and use its size - offset as content length instead of just using its size.

Other possible alternatives:

• check if the offset into the file is not 0 after operning it, and perform a chunked upload in that case

$ nc -l 43000
PUT /4 HTTP/1.1
Host: localhost:43000
User-Agent: curl/8.2.1
Accept: */*
Transfer-Encoding: chunked
Expect: 100-continue

1d
Tom     Carpenter
Alice   Mechanic

0
[...]

• seek to offset 0 after opening the regular file and send the entire file:

$ nc -l 43000
PUT /4 HTTP/1.1
Host: localhost:43000
User-Agent: curl/8.2.1
Accept: */*
Content-Length: 38

NAME    JOB
Tom     Carpenter
Alice   Mechanic

Those solutions are all fairly easy to implement. I, personally, would prefer either the first or the second approach, I don't really like the idea of curl seeking back to 0 after opening.

curl/libcurl version

curl 8.2.1 (I am using my https://sdf.org/ shell account to test this)

curl 8.2.1 (x86_64--netbsd) libcurl/8.2.1 OpenSSL/1.1.1u zlib/1.2.13 libidn2/2.3.4 nghttp2/1.52.0
Release-Date: 2023-07-26
Protocols: dict file ftp ftps gopher gophers http https imap imaps mqtt pop3 pop3s rtsp smb smbs smtp smtps telnet tftp
Features: alt-svc AsynchDNS GSS-API HSTS HTTP2 HTTPS-proxy IDN IPv6 Kerberos Largefile libz NTLM NTLM_WB SPNEGO SSL threadsafe TLS-SRP UnixSocket

operating system

NetBSD sdf 9.3 NetBSD 9.3 (GENERIC) #0: Thu Aug  4 15:30:37 UTC 2022  [email protected]:/usr/src/sys/arch/amd64/compile/GENERIC amd64