+2007-07-23 Ben Pfaff <blp@gnu.org>
+
+ Improvements to system file reader and writer.
+
+ First, move all detailed knowledge of very long strings into
+ sys-file-private.[ch], so that this nasty stuff can be isolated.
+
+ * sys-file-private.c (REAL_VLS_CHUNK): New macro.
+ (EFFECTIVE_VLS_CHUNK): New macro.
+ (min_int): New function.
+ (max_int): New function.
+ (sfm_width_to_bytes): Rewrite.
+ (sfm_width_to_octs): New function.
+ (sfm_segment_alloc_width): New function.
+ (sfm_segment_alloc_bytes): New function.
+ (sfm_segment_used_bytes): New function.
+ (sfm_segment_offset): New function.
+ (sfm_segment_effective_offset): New function.
+ (sfm_dictionary_to_sfm_vars): New function.
+
+ * sys-file-private.h (MIN_VERY_LONG_STRING): Removed.
+ (EFFECTIVE_LONG_STRING_LENGTH): Removed.
+ (struct sfm_var): New structure.
+
+ Next, improvements to the system file reader.
+
+ * sys-file-reader.h (struct sfm_read_info): Changed `case_cnt' to
+ type casenumber. Added `version_major', `version_minor',
+ `version_revision'.
+
+ * sys-file-reader.c (struct sfm_reader): Replaced `flt64_cnt' by
+ `oct_cnt'. Rename `vars', `var_cnt' to `sfm_vars', `sfm_var_cnt'.
+ Change `case_cnt' to type casenumber. Removed `has_vls'.
+ (struct sfm_var): Removed.
+ (sfm_open_reader): Don't warn on wrong case size if the file was
+ written by SPSS 13, which tends to get it wrong. Use
+ sfm_dictionary_to_sfm_vars.
+ (read_header): Always output system file info.
+ (read_variable_record): Simplify code for reading missing values.
+ (read_machine_int32_info): Save version numbers from system file
+ into info struct passed as new argument.
+ (read_long_string_map): Restructured to use new sys-file-private
+ functions.
+ (read_value_labels): Use size_overflow_p.
+ (sys_file_casereader_read): Get rid of distinction between fast
+ and slow paths. Use information provided by sys-file-primate's
+ struct sfm_var to simplify code.
+ (skip_whole_strings): New function.
+ (read_int32): Renamed read_int. Changed return value to int.
+ Updated all callers.
+ (read_flt64): Renamed read_float. Changed return value to
+ double. Updated all callers.
+ (int32_to_native): Removed. Changed callers to use
+ integer_convert.
+ (flt64_to_double): Removed. Changed callers to use float_convert.
+
+ Finally, get rid of int32, flt64 terminology and types in system
+ file writer. The former wasn't very useful since a POSIX "int"
+ can hold the whole range of int32 and we generally didn't have a
+ need for it to be exactly-32-bits, just at-least-32-bits. The
+ latter was inconvenient because we had to assume that it could be
+ different from double and thereby convert special values SYSMIS,
+ HIGHEST, LOWEST to and from it in multiple places. Instead, now
+ we just use "int" and "double" in most places, and do conversions,
+ if necessary, very close to where we do I/O. This change meant
+ that the writer code couldn't represent records in the file as C
+ structs any longer, but that's no great loss. The code actually
+ seems to be more readable without them.
+
+ Simplify the compression buffering code: only buffer as much as
+ necessary, which is no more than eight 8-byte units at any given
+ time.
+
+ * sys-file-writer.c (typedef flt64): Removed.
+ (macro second_lowest_flt64): Removed.
+ (struct sysfile_header): Removed.
+ (struct sysfile_variable): Removed.
+ (struct sfm_writer): Removed `needs_translation', `has_vls',
+ `flt64_cnt'. Changed `compress' to type bool and `case_cnt' to
+ type casenumber. Renamed `vars' to `sfm_vars', `var_cnt' to
+ `sfm_var_cnt'. Replaced `buf', `end', `ptr', `x', `y' for
+ compression buffering by `opcodes', `opcode_cnt', `data',
+ `data_cnt'. Renamed `var_cnt_vls' as `segment_cnt'.
+ (sfm_open_writer): Use sfm_dictionary_to_sfm_vars. Use simple
+ data writer functions instead of structures.
+ (calc_oct_idx): New function.
+ (write_header): Use simple data writer functions instead of
+ structures.
+ (write_format_spec): Renamed write_format. New argument.
+ (write_variable_continuation_records): New function.
+ (write_variable): Use simple data writer functions instead of
+ structures. Use write_variable_continuation_records. Write
+ entire very long string instead of requiring caller to understand
+ them.
+ (write_value_labels): Use simple data writer functions instead of
+ structures.
+ (write_documents): Ditto.
+ (write_variable_display_parameters): Use sys-file-private
+ functions to simplify. Use simple data writer functions instead
+ of structures.
+ (write_vls_length_table): Use simple data writer functions instead
+ of structures.
+ (write_longvar_table): Ditto.
+ (write_rec_7_34): Break into new functions
+ write_integer_info_record, write_float_info_record. Use simple
+ data writer functions instead of structures.
+ (buf_write): Removed.
+ (append_string_max): Removed.
+ (ensure_buf_space): Removed.
+ (sys_file_casewriter_write): Get rid of the distinction between
+ fast and slow paths, which didn't seem to be too useful. Use new
+ functions write_case_uncompressed, write_case_compressed.
+ (put_instruction): Removed.
+ (put_element): Removed.
+ (write_compressed_data): Removed.
+ (close_writer): Use flush_compressed. Only write case count to
+ system file if it will fit in the field.
+ (write_case_compressed): New function.
+ (write_case_uncompressed): New function.
+ (flush_compressed): New function.
+ (put_cmp_opcode): New function.
+ (put_cmp_number): New function.
+ (write_int): New function.
+ (convert_double_to_output_format): New function.
+ (write_float): New function.
+ (write_value): New function.
+ (write_string): New function.
+ (write_bytes): New function.
+ (write_zeros): New function.
+ (write_spaces): New function.
+
2007-07-22 Ben Pfaff <blp@gnu.org>
Don't try to write very long strings to portable files. The
projects / pspp-builds.git / blobdiff