[HADOOP-1758] processing escapes in a jute record is quadratic

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 0.13.0
• Fix Version/s: 0.15.0
• Component/s: record
• Assignee: Vivek Ratan
• Reporter: Dick King
• Created: 22/Aug/07 18:03
• Updated: 05/Nov/07 18:12
• Resolved: 05/Sep/07 21:56

Description

The following code appears in hadoop/src/c++/librecordio/csvarchive.cc :

static void replaceAll(std::string s, const char *src, char c)
{
std::string::size_type pos = 0;
while (pos != std::string::npos) {
pos = s.find(src);
if (pos != std::string::npos)

{ s.replace(pos, strlen(src), 1, c); }

}
}

This is used in the context of replacing jute escapes in the code:

void hadoop::ICsvArchive::deserialize(std::string& t, const char* tag)
{
t = readUptoTerminator(stream);
if (t[0] != '\'')

{ throw new IOException("Errror deserializing string."); }

t.erase(0, 1); /// erase first character
replaceAll(t, "%0D", 0x0D);
replaceAll(t, "%0A", 0x0A);
replaceAll(t, "%7D", 0x7D);
replaceAll(t, "%00", 0x00);
replaceAll(t, "%2C", 0x2C);
replaceAll(t, "%25", 0x25);

}

Since this replaces the entire string for each instance of the escape sequence, practically anything would be better. I would propose that within deserialize we allocate a char * [since each replacement is smaller than the original], scan for each %, and either do a general hex conversion in place or look for one of the six patterns, and after each replacement move down the unmodified text and scan for the % fom that starting point.

-dk