Previously, MultipartEncoder in conjunction with requests could produce a malformed request if the source data stream grew (e.g. appending to a log file being uploaded) or shrunk during encoding. This change addresses both the growing case (by only transmitting the originally promised amount of data) and the shrinking case (by raising an exception).
Note that the data produced by MultipartEncoder itself was well formed. The issue occurs when downstream code makes promises or assumptions based on the initial len (e.g. transmitting it to the server as Content-Length, as requests does), only to have the encoder return more/less data. This change may break (presumably rare) use cases that rely on both a) on shrinking/growing the source data stream during encoding, and b) don't already fail due to the promised/actual encoded data length mismatch.
Previously, MultipartEncoder in conjunction with requests could produce a malformed request if the source data stream grew (e.g. appending to a log file being uploaded) or shrunk during encoding. This change addresses both the growing case (by only transmitting the originally promised amount of data) and the shrinking case (by raising an exception).
Note that the data produced by MultipartEncoder itself was well formed. The issue occurs when downstream code makes promises or assumptions based on the initial
len(e.g. transmitting it to the server asContent-Length, as requests does), only to have the encoder return more/less data. This change may break (presumably rare) use cases that rely on both a) on shrinking/growing the source data stream during encoding, and b) don't already fail due to the promised/actual encoded data length mismatch.