Saxon » SaxonJS

For the "q" flag case, the conversion should just be replacing all $ by $$, i.e. str.replace(/\$/g, "$$$$").

In the other case, the new code splits up the conversion steps, and adds details to ensure conversions happen correctly (i.e. that the regex string is parsed correctly e.g. the string \\$0 contains the special characters \\ and $0, not the character \$).

Step 1: $n leave as is except replace $0 by $& str = str.replace(/(^|[^\\])(\\\\)*\$0/g, "$1$2$$\&");

Step 2: for $ not followed by digit, replace \$ by $$ str = str.replace(/(^|[^\\])(\\\\)*\\\$([^0-9]|[^&]|$)/g, "$1$2$$$$$3");

Step 3: replace \\ by \ str = str.replace(/\\\\/g, "\\");

Code changes committed on 1.x and trunk branches.

Reopening because the patch is insufficient. As demonstrated by the failure of QT3 test fn-replace-45, which tests the following:

replace("Now, let's SEND OUT for QUICHE!!", "[A-Z][A-Z]+", "$0$0")

The problem here is that the replacement string has consecutive occurrences of $0, which are not handled correctly by step 1. The convertReplacement function returns $&$0 rather than $&$&.

Similarly, consecutive occurrences of \$ will not be handled correctly by step 2.

Using the Javascript String.replace() function to convert the replacement string was becoming too involved and confusing. Instead, the code in convertReplacement has been changed to parse and convert the replacement string directly using a switch on characters in the string.

The regex check for invalid replacement strings /(^|[^\\])\$([^0-9]|$)/.test(str) || /(^|[^\\])\\([^\\\$]|$)/.test(str) is retained, to filter out invalid strings. This means that within the switch it can be assumed that the replacement string is valid. So certain conditions can be assumed - e.g. an unescaped dollar must be followed by a digit; and an unescaped backslash must be followed by a backslash or dollar.

Patch committed on Saxon-JS 1 and trunk branches.

Bug fix applied in the Saxon-JS 1.2.0 maintenance release.