1 All about co_lnotab, the line number table. 2 3 Code objects store a field named co_lnotab. This is an array of unsigned bytes 4 disguised as a Python bytes object. It is used to map bytecode offsets to 5 source code line #s for tracebacks and to identify line number boundaries for 6 line tracing. 7 8 The array is conceptually a compressed list of 9 (bytecode offset increment, line number increment) 10 pairs. The details are important and delicate, best illustrated by example: 11 12 byte code offset source code line number 13 0 1 14 6 2 15 50 7 16 350 207 17 361 208 18 19 Instead of storing these numbers literally, we compress the list by storing only 20 the difference from one row to the next. Conceptually, the stored list might 21 look like: 22 23 0, 1, 6, 1, 44, 5, 300, 200, 11, 1 24 25 The above doesn't really work, but it's a start. An unsigned byte (byte code 26 offset) can't hold negative values, or values larger than 255, a signed byte 27 (line number) can't hold values larger than 127 or less than -128, and the 28 above example contains two such values. (Note that before 3.6, line number 29 was also encoded by an unsigned byte.) So we make two tweaks: 30 31 (a) there's a deep assumption that byte code offsets increase monotonically, 32 and 33 (b) if byte code offset jumps by more than 255 from one row to the next, or if 34 source code line number jumps by more than 127 or less than -128 from one row 35 to the next, more than one pair is written to the table. In case #b, 36 there's no way to know from looking at the table later how many were written. 37 That's the delicate part. A user of co_lnotab desiring to find the source 38 line number corresponding to a bytecode address A should do something like 39 this: 40 41 lineno = addr = 0 42 for addr_incr, line_incr in co_lnotab: 43 addr += addr_incr 44 if addr > A: 45 return lineno 46 if line_incr >= 0x80: 47 line_incr -= 0x100 48 lineno += line_incr 49 50 (In C, this is implemented by PyCode_Addr2Line().) In order for this to work, 51 when the addr field increments by more than 255, the line # increment in each 52 pair generated must be 0 until the remaining addr increment is < 256. So, in 53 the example above, assemble_lnotab in compile.c should not (as was actually done 54 until 2.2) expand 300, 200 to 55 255, 255, 45, 45, 56 but to 57 255, 0, 45, 127, 0, 73. 58 59 The above is sufficient to reconstruct line numbers for tracebacks, but not for 60 line tracing. Tracing is handled by PyCode_CheckLineNumber() in codeobject.c 61 and maybe_call_line_trace() in ceval.c. 62 63 *** Tracing *** 64 65 To a first approximation, we want to call the tracing function when the line 66 number of the current instruction changes. Re-computing the current line for 67 every instruction is a little slow, though, so each time we compute the line 68 number we save the bytecode indices where it's valid: 69 70 *instr_lb <= frame->f_lasti < *instr_ub 71 72 is true so long as execution does not change lines. That is, *instr_lb holds 73 the first bytecode index of the current line, and *instr_ub holds the first 74 bytecode index of the next line. As long as the above expression is true, 75 maybe_call_line_trace() does not need to call PyCode_CheckLineNumber(). Note 76 that the same line may appear multiple times in the lnotab, either because the 77 bytecode jumped more than 255 indices between line number changes or because 78 the compiler inserted the same line twice. Even in that case, *instr_ub holds 79 the first index of the next line. 80 81 However, we don't *always* want to call the line trace function when the above 82 test fails. 83 84 Consider this code: 85 86 1: def f(a): 87 2: while a: 88 3: print(1) 89 4: break 90 5: else: 91 6: print(2) 92 93 which compiles to this: 94 95 2 0 SETUP_LOOP 26 (to 28) 96 >> 2 LOAD_FAST 0 (a) 97 4 POP_JUMP_IF_FALSE 18 98 99 3 6 LOAD_GLOBAL 0 (print) 100 8 LOAD_CONST 1 (1) 101 10 CALL_FUNCTION 1 102 12 POP_TOP 103 104 4 14 BREAK_LOOP 105 16 JUMP_ABSOLUTE 2 106 >> 18 POP_BLOCK 107 108 6 20 LOAD_GLOBAL 0 (print) 109 22 LOAD_CONST 2 (2) 110 24 CALL_FUNCTION 1 111 26 POP_TOP 112 >> 28 LOAD_CONST 0 (None) 113 30 RETURN_VALUE 114 115 If 'a' is false, execution will jump to the POP_BLOCK instruction at offset 18 116 and the co_lnotab will claim that execution has moved to line 4, which is wrong. 117 In this case, we could instead associate the POP_BLOCK with line 5, but that 118 would break jumps around loops without else clauses. 119 120 We fix this by only calling the line trace function for a forward jump if the 121 co_lnotab indicates we have jumped to the *start* of a line, i.e. if the current 122 instruction offset matches the offset given for the start of a line by the 123 co_lnotab. For backward jumps, however, we always call the line trace function, 124 which lets a debugger stop on every evaluation of a loop guard (which usually 125 won't be the first opcode in a line). 126 127 Why do we set f_lineno when tracing, and only just before calling the trace 128 function? Well, consider the code above when 'a' is true. If stepping through 129 this with 'n' in pdb, you would stop at line 1 with a "call" type event, then 130 line events on lines 2, 3, and 4, then a "return" type event -- but because the 131 code for the return actually falls in the range of the "line 6" opcodes, you 132 would be shown line 6 during this event. This is a change from the behaviour in 133 2.2 and before, and I've found it confusing in practice. By setting and using 134 f_lineno when tracing, one can report a line number different from that 135 suggested by f_lasti on this one occasion where it's desirable. 136
