Search and Replace - possible bug?

[PL_kolek]

Hello!
From the information on this forum:

Note that the search & replace operates on the XHTML Calibre produces during conversion, not on the original file.

But I think that is not true. I have a TXT file with a string "* * *" being some kind of a chapter beginning. When I choose search and replace for this file, I can see the XHTML file. The stars are missing (another problem), but I have string "<i> </i>" in the same places. So I fill the search for field with:

<i> </i>

,
get around 70 hits, and replace it with, let's say 123456. Then I convert, and there is no such string in the output file, so no replaces took place, despite those hits.

However, when I make such search:

\* \* \*

,
get no hits at all, but all 70 entries in the source txt (not xhtml in the middle of the conversion) get substituted.

For that seems like a huge bug, or at least not predictable behaviour. When I see hits, I expect my substitutions to really work. When I get no hits, there should be no replacements. I think the order of conversion is wrong. FIt should first make it xhtml and then apply s&r, not the other way. Or (I would really love if that would work that way) we could specify 2 searches, one for the source file and the second for the middle xhtml

When I searched that forum for topics according to this functionality (s&r), I found few people havin the same problem as I had (not the first time, though). The results were unpredictable and replacing didn't work at all.

[kovidgoyal]

The way txt input is interpreted is heavily dependent on heuristics and conversion options. Convert to a more stable format like epub of htmlz first, then reconvert that and run S&R on it.

[PL_kolek]

But in my opinion the order of conversion shouldn't depend on format. If I see something in the wizard, and then it isn't applied to my e-book just because Calibre in the background uses it in the other phase, then something is wrong.

Thanks for the tip, I'll try that also.