Question

I’m struggling to create a sed regex command to change lines like this:

In file included from dira/file_a.h:8, dire/file_e.h:9, and dirf/file_f.h:10,
             from dirb/file_b.h:6,
             from /existing/abs/path/dirb/file_b.cc:6:
dirc/file_c.h:88: error: 'eqn_count_t' does not name a type
dirc/file_c.h:95: error: 'wave_count_t' does not name a type
dirc/file_c.h:104: error: ISO C++ forbids declaration of 'WmHyperbolicEqnSet' with no type

To this desired output:

In file included from /abspaths/dira/file_a.h:8, /abspaths/dire/file_e.h:9, and /abspaths/dirf/file_f.h:10,
             from /abspaths/dirb/file_b.h:6,
             from /existing/abs/path/dirb/file_b.cc:6:
/abspaths/dirc/file_c.h:88: error: 'eqn_count_t' does not name a type
/abspaths/dirc/file_c.h:95: error: 'wave_count_t' does not name a type
/abspaths/dirc/file_c.h:104: error: ISO C++ forbids declaration of 'WmHyperbolicEqnSet' with no type

so,

• only match relative path+filenames that end in .h
• do not match lines that start with a forward-slash (and thus are already absolute paths)
• match multiple occurances per line
• It has become apparent that I need a command that works with Mac OS X‘s BSD sed command.

What is the regex and sed command that I want?

I’m trying to modify gcc output because included header files with errors/warnings generate error stream output with the relative path referenced, not absolute path. With my XCode IDE calling an external build system, the errors occurring in .h files are not ‘clickable’.

Answer 1

Mac and Linux friendly:

sed -E 's/^([^/][a-zA-Z/_]+\.h)/\/abspaths\/\1/;s/ ([^/][a-zA-Z/_]+\.h)/ \/abspaths\/\1/g' file

Matches desired output:

In file included from /abspaths/dira/file_a.h:8, /abspaths/dire/file_e.h:9, and /abspaths/dirf/file_f.h:10,
             from /abspaths/dirb/file_b.h:6,
             from /existing/abs/path/dirb/file_b.cc:6:
/abspaths/dirc/file_c.h:88: error: 'eqn_count_t' does not name a type
/abspaths/dirc/file_c.h:95: error: 'wave_count_t' does not name a type
/abspaths/dirc/file_c.h:104: error: ISO C++ forbids declaration of 'WmHyperbolicEqnSet' with no type

Explanation:

Two substitution are required to account for the extra space needed when the replacement isn’t at the start of the line:

s/^([^/][a-zA-Z/_]+\.h)/\/abspaths\/\1/;   # First substitution for start of line 
s/ ([^/][a-zA-Z/_]+\.h)/ \/abspaths\/\1/g  # Second for non-start of line

# Match (first substitution)
s/
^             - start of line
(             - capture group 
[^/]          - not a forward slash 
[a-zA-Z/_]+   - one or more letter, forward slash or underscore
\.h           - the extension (escaped) 
)             - end capture group 
# Replace with 
/
\/abspaths\/  - the literal string /abspaths (slashes escaped)
\1            - the captured group 
/;
# Match (second substitution)
s/
' '           - not start of line but a single space (used quotes here for space)
(             - capture group 
[^/]          - not a forward slash 
[a-zA-Z/_]+   - one or more letter, forward slash or underscore
\.h           - the extension (escaped) 
)             - end capture group 
# Replace with 
/
' '           - put the single space back
\/abspaths\/  - the literal string /abspaths (slashes escaped)
\1            - the captured group 
/g            - global flag

---

Or just by doing one substitution (based on F.Hauri) answer however will only work for one match per line:

sed -E 's/^(.* )?([^/][^ ]+\.h)/\1\/abspath\/\2/' file

For multiple matches sed supports branching:

sed -E ':a;s/^(.* )?([^/][^ ]+\.h)/\1\/abspath\/\2/;ta' file