| カテゴリ | 重要度 | ステータス | 解決状況 | 登録日時 | 最終更新 |
|---|---|---|---|---|---|
 システムコア | 低 | 認証済み | 未処理 | 2013-08-26 02:15 | 2013-08-26 10:21 |
| テスター | mizapf | 担当者 | ソース | ||
| バージョン | 0.149u1 | 発生バージョン | 修正バージョン | ||
| 修正コミット | プルリク | ||||
| フラグ | |||||
| セット | All systems using sdlsocket | ||||
| セット詳細 | |||||
| 概 要 | sdlsocketが入力バッファを失う。 | ||||
| 詳 細 | The socket is designed as a special sdlfile which allows for input and output operations. That is, you can perform read and write operations on that sdlfile. However, once you do a core_fwrite, the input buffer will be flushed (file->bufferbytes = 0), and all data in the input buffer that have not been read at that point of time are lost. This is not the expected behavior of a socket which is supposed to have independent input and output queues. | ||||
| 再現手順 | |||||
| 追加情報 | I noticed this problem when trying to find out why the RS232 connections are not reliable that I implemented some time ago in MESS/ti99_4x. I'm using a socket connection to an external tool that interfaces with the PC UART. I assumed that it was safe to do repeated fread operations with 1 byte, keeping the received data in the receive buffer of the UART. I did not expect that a write operation (for handshake or echo) wipes the buffer. One workaround is that I include a private buffer in my serial adapter emulation so that I quickly pull out all data from the sdlfile before a write occurs. We could also think about ways to fix the sdlsocket implementation because this will become a general issue whereever a socket is used for bidirectional communication. Added 08-14-2015: "..found a workaround, as described there (see ti_rs232.h:81) which is working well enough. The problem (cleaning the buffer) is still there." | ||||
| 添付ファイル | |||||