[fasharif]

What was wrong

• Memory was overstated by thousands of times. After stime, the parser skipped 7 fields and read field 23 of /proc/[pid]/stat (vsize, the virtual size in bytes) as the resident set size. It then multiplied that by a hard-coded 4 and labelled it KB. For a shell using under 1 MB of RAM, it would print 6,766,592 KB.
• CPU% used the wrong formula. It subtracted the process's CPU time from the system uptime, instead of the process's start time. Even the corrected formula only gives a lifetime average, not current use like top.
• Names containing ) broke parsing, because the name was read with %255[^)].

What changed

• Memory now comes from field 24 (rss, in pages) multiplied by sysconf(_SC_PAGESIZE).
• CPU% compares two readings of utime + stime taken -i seconds apart, as top does.
• The name is taken from the first ( to the last ), and every file read and parse is checked.
• New options: -i (sampling interval), -n (row limit) and -s cpu|mem (sort order).
• Parsing and calculations moved into src/proc.c so they can be unit tested.

How it's tested

• Unit tests (make test): a real /proc/1/stat line, which asserts that memory comes from field 24 and not field 23; names with spaces and parentheses; overlong names; malformed lines; and the CPU% and memory calculations.
• End-to-end (make integration): runs the program on a fake /proc tree, compares its memory figure for PID 1 with ps on the live system, and checks that invalid options exit with status 2.
• CI: builds with GCC and Clang with warnings as errors, runs both suites, repeats them under AddressSanitizer and UBSan, and runs cppcheck.

🤖 Generated with Claude Code