| Summary: | the sed applet behaves erraticaly when script from file (option -f) | ||
|---|---|---|---|
| Product: | Busybox | Reporter: | clu <clu> |
| Component: | Standard Compliance | Assignee: | unassigned |
| Status: | RESOLVED INVALID | ||
| Severity: | normal | CC: | busybox-cvs |
| Priority: | P5 | ||
| Version: | 1.22.x | ||
| Target Milestone: | --- | ||
| Hardware: | PC | ||
| OS: | Windows | ||
| Host: | Target: | ||
| Build: | |||
| Attachments: | tarball of the web extraction used as sample to reproduce the bug | ||
(In reply to comment #0) > Created attachment 5240 [details] > tarball of the web extraction used as sample to reproduce the bug > > context (used as bash for the shell and sed for the operations) : > BusyBox v1.23.0-TIG-1381-g883817e (2014-01-21 14:06:02 GMT) multi-call binary. > > i perform a succession of transformations with bb's sed on a given file (see > attached for a sample). > > 1. on the command line : > $ cat samp.htm | sed -r "/sens=achat/! d; s/^[^<]*<//g; s/<tr /\n&/g" | sed > -r "s/.*symbole=([^']*)' title=([^>]*)[^<]*<\/a><\/td>/\"\1\";\2;/g" > works as expected. > > 2. and i want to do the same via a sed file : > $ cat samp.htm | sed -rf f2.sed > where f2. contains the following (without the line numbering): > 1- /sens=achat/! d > 2- s/^[^<]*<//g; s/<tr /\n&/g > 3- s/.*symbole=([^']*)' title=([^>]*)[^<]*<\/a><\/td>/\"\1\";\2;/g > but it seems that the script doesn't go beyond line 2 (as if, for the cli > version, the last pipe was skipped!) (You need to stop escaping \" in the file version. IOW: \"\1\" should be "\1" there). For me the script doesn't work as if line 3 is missing. The output is different from that too. However: without going deeper into what the scripts are doing, I tried GNU sed version 4.2.1 on them too, and it and busybox sed gave exactly the same results on both your testcases. Moreover, sed -r \ -e "/sens=achat/! d; s/^[^<]*<//g; s/<tr /\n&/g" \ -e "s/.*symbole=([^']*)' title=([^>]*)[^<]*<\/a><\/td>/\"\1\";\2;/g" that is, single process instead of two seds piped, works the same as the version with -f FILE. I guess you are inserting \n chars to break lines and expect sed to detect those as new lines to operate one. This doesn't work. Thank you for taking care of my problem. > (You need to stop escaping \" in the file version. IOW: \"\1\" should be "\1" there). My bad, i didn't notice... you totally right > For me the script doesn't work as if line 3 is missing. The output is different from that too. That is precisely my point but ... ... > I guess you are inserting \n chars to break lines and expect sed to detect those as new lines to operate one. This doesn't work. I understand here that, the 3 operations are realized on the same original file while for the piped version last two operations are made on the "piped results" of the previous ones... thus even : sed -r -e "/sens=achat/! d; s/^[^<]*<//g; s/<tr /\n&/g" is different from : sed -r -e "/sens=achat/! d" | sed -r -e "s/^[^<]*<//g; s/<tr /\n&/g" because in the first case the substitutions would also be done on the (later) deleted lines whereas in the second case they would only be applied to the remaining lines, after deletion. Am i understanding right? If so, that would explain to me why the commands can't be gathered in a single sed script file. Again, many thank for your help. |
Created attachment 5240 [details] tarball of the web extraction used as sample to reproduce the bug context (used as bash for the shell and sed for the operations) : BusyBox v1.23.0-TIG-1381-g883817e (2014-01-21 14:06:02 GMT) multi-call binary. i perform a succession of transformations with bb's sed on a given file (see attached for a sample). 1. on the command line : $ cat samp.htm | sed -r "/sens=achat/! d; s/^[^<]*<//g; s/<tr /\n&/g" | sed -r "s/.*symbole=([^']*)' title=([^>]*)[^<]*<\/a><\/td>/\"\1\";\2;/g" works as expected. 2. and i want to do the same via a sed file : $ cat samp.htm | sed -rf f2.sed where f2. contains the following (without the line numbering): 1- /sens=achat/! d 2- s/^[^<]*<//g; s/<tr /\n&/g 3- s/.*symbole=([^']*)' title=([^>]*)[^<]*<\/a><\/td>/\"\1\";\2;/g but it seems that the script doesn't go beyond line 2 (as if, for the cli version, the last pipe was skipped!) For the story this is a web extraction from boursorama.com (collected with bb's wget) which i'm transforming into csv (with ";" as separator -french version...-) I am available for further tests on my "win XP pro" platform thank you for your concern. clu 20140214