|
2026-10-08
| ||
| 21:57 | • New ticket [04b8e4fb3c] buildmessage/copymessage emit Content-Transfer-Encoding header twice. artifact: 6cd8021cd7 user: anonymous | |
| Ticket UUID: | 04b8e4fb3cc43f9b88065af15e789a83bf279957 | |||
| Title: | buildmessage/copymessage emit Content-Transfer-Encoding header twice | |||
| Type: | Bug | Version: | 1.7.2 | |
| Submitter: | anonymous | Created on: | 2026-10-08 21:57:27 | |
| Subsystem: | mime | Assigned To: | nobody | |
| Priority: | 5 Medium | Severity: | Important | |
| Status: | Open | Last Modified: | 2026-10-08 21:57:27 | |
| Resolution: | None | Closed By: | nobody | |
| Closed on: | ||||
| Description: |
For any canonical, non-multipart part, mime::buildmessage and mime::copymessage write the Content-Transfer-Encoding header twice:
Output:
This happens whether or not -encoding is given, because both code paths fall back to [mime::encoding $token]. RFC 2045 allows at most one Content-Transfer-Encoding field per part. AWS SES (SendEmail, Raw content) rejects these messages with a duplicate-header error. Cause: Check-in 7c25c9a457 ("Separate header determination from body generation", 2021-04-30) changed mime::getheader (called without a key) to add Content-Transfer-Encoding to its result through mime::getTransferEncoding. mime::copymessageaux prints every header returned by getheader, then still runs its older code that writes the header again:
Suggested fix: Remove that puts from mime::copymessageaux. Keep the switch after it, because it still has to set $converter for base64/quoted-printable bodies. Add a test asserting that [mime::buildmessage] output contains exactly one Content-Transfer-Encoding line for -encoding 8bit, base64 and quoted-printable, and when -encoding is omitted. | |||
