Fix Unicode character corruption and lack of XML character escaping

[zlondrej]

The originally used xshowBlob function converted String to ByteString by truncating the character to 8-bit value, essentially limiting the character set to Latin 1.

It also didn't escape XML characters properly.

[zlondrej]

@claude review

[arybczak]

Looks good, though a small regression test and a helper function for reducing copy-paste would indeed be nice, as Claude noted.

Review of commit 62e44d7

Commit: "Fix Unicode character corruption and lack of XML character escaping"

Verdict

Approve. The change is correct.

What the commit changes

The commit replaces xshowBlob with HXT.xshowEscapeXml in three places.
It encodes the result as UTF-8 with Data.ByteString.Lazy.UTF8.fromString.

• SAML2/XML.hs: docToXMLWithoutRoot and docToXMLWithRoot
• SAML2/XML/Signature.hs: the fallback case of applyTransformsXML

Why the old code was wrong

xshowBlob in hxt-9.3.1.22 calls stringToBlob, which is
Data.ByteString.Lazy.Char8.pack. That function keeps the low 8 bits of each
character. Every character outside Latin-1 became a different byte.

xshowBlob also calls xshow, which passes text and attribute values through
showString. That function does not escape any character. A text node with
& or < produced malformed XML.

Why the new code is right

xshowEscapeXml is defined in Text.XML.HXT.Arrow.Edit as
f >. uncurry xshow'' escapeXmlRefs. The escapeXmlRefs pair works as
follows:

• In text content, cquote escapes < and &.
• In attribute values, aquote escapes <, >, ", ', &, newline, tab
  and carriage return.

The >. combinator collects all results into one list, so the arrow yields
exactly one String. The concat after runLA is safe.

Verification

• The test suite passes on GHC 9.10.3: 25 cases, 0 errors, 0 failures.

• A round trip through docToXMLWithRoot and xmlToDoc returns an
  identical tree for this input:

  <a x="q&quot;&lt;&#10;">Zażółć gęślą jaźń &amp; &lt;b&gt; 日本</a>

Notes, none blocking

1. The commit adds no regression test. Add a round-trip case to test/XML.hs
   with a non-ASCII string and an ampersand. This guards both bugs.
2. The expression BSLU.fromString . concat . HXT.runLA (HXT.xshowEscapeXml ...)
   appears three times. One helper in SAML2.XML that takes the selector
   arrow removes the duplication. SAML2.XML.Signature can then drop its
   Data.ByteString.Lazy.UTF8 import.
3. xshowEscapeXml does not escape > in text content. A text node that
   contains the sequence ]]> serializes to output that xmlToDocE rejects
   with unexpected ']'. This is an HXT limitation, not a regression. The
   sequence is unlikely in SAML payloads.

[zlondrej]

Regression tests and encoding function added.