In text files, ftell() produces only encoded offsets. It returns a long int, in which the block number and the byte offset within the block are encoded. You cannot rely on any encoded offset not returned by ftell() except 0. This includes encoded offsets that you adjust yourself (for example, with addition or subtraction).
When you call fseek() with the whence value SEEK_SET, you must use an encoded offset returned from ftell(). For whence values of SEEK_CUR and SEEK_END, however, you specify relative byte offsets. If you want to seek to a certain relative byte offset, you can use SEEK_SET with an offset of 0 to rewind the file to the beginning, and then you can use SEEK_CUR to specify the desired relative byte offset. z/OS® XL C/C++ counts new-line characters and skips to the next record each time it reads one.
Unlike binary files you cannot specify offsets for SEEK_CUR and SEEK_END that set the file position past the end of the file. Any offset that indicates a position outside the current file is invalid and causes fseek() to fail.
In earlier releases, ftell() could determine position only for files with no more than 131071 blocks. In the new design, this number increases depending on the block size. From a maximum block size of 32760, every time this number decreases by half, the number of blocks that can be represented doubles. Using the large file version of ftello() for a large format sequential data set increases the maximum number of blocks that can be represented to 2 GB in AMODE 31 C/C++ applications.
Repositioning flushes all updates before changing position. An invalid call to fseek() is now always treated as a flush. It flushes all updated records or all complete new records in the block, and leaves the file position unchanged. If the flush fails, any characters in the ungetc() buffer are lost. If a block contains an incomplete new record, the block is saved and will be completed by another write or by closing the file.