Demo: Cygwin "eats up" backslashes unexpectedly under certain circumstances

cygwin-eats-backslashes.c

#include <ctype.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <wchar.h>

// https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/glob.cc#L112
#define MAXPATHLEN 8192

// https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/glob.cc#L112
#define EOS '\0'
#define QUOTE '\\'

// https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/glob.cc#L135
#define M_PROTECT 0x4000000000ULL

// https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/glob.cc#L148
typedef char Char;

// https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/dcrt0.cc#L147-L152
static inline int isquote(char c)
{
    char ch = c;
    return ch == '"' || ch == '\'';
}

// https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/glob.cc#L187-L254
// The code is hugely modified based on the calling signature of https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/dcrt0.cc#L265
char *glob_partial(char *pattern)
{
    const char *patnext;
    Char *bufnext, *bufend, *patbuf = malloc(MAXPATHLEN), prot;
    mbstate_t mbs;
    wchar_t wc;
    size_t clen;

    patnext = pattern;
    bufnext = patbuf;
    bufend = bufnext + MAXPATHLEN - 1;

    memset(&mbs, 0, sizeof(mbs));
    while (bufend - bufnext >= MB_CUR_MAX)
    {
        if (*patnext == QUOTE)
        {
            if (*++patnext == EOS)
            {
                *bufnext++ = QUOTE | M_PROTECT;
                continue;
            }
            prot = M_PROTECT;
        }
        else
            prot = 0;
        clen = mbrtowc(&wc, patnext, MB_LEN_MAX, &mbs);
        if (clen == (size_t)-1 || clen == (size_t)-2)
            return malloc(0);
        else if (clen == 0)
            break;
        *bufnext++ = wc | prot;
        patnext += clen;
    }
    *bufnext = EOS;

    return patbuf;
}

// https://github.com/cygwin/cygwin/blob/eeeb5650cf706f4dde72ce8b8598aef41f88718a/winsup/cygwin/dcrt0.cc#L205-L288
// The code is hugely modified to illustrate the interest of this test code
void globify_test(char *word)
{
    int n = 0;
    char *p, *s;
    int dos_spec = 0;

    /* We'll need more space if there are quoting characters in
     word.  If that is the case, doubling the size of the
     string should provide more than enough space. */
    if (strpbrk(word, "'\""))
        n = strlen(word);
    char pattern[strlen(word) + ((dos_spec + 1) * n) + 1];

    printf("original input: %s\n", word);

    for (p = pattern, s = word; *s != '\000'; s++, p++)
    {
        if (!isquote(*s))
        {
            if (dos_spec && *s == '\\')
                *p++ = '\\';
            *p = *s;
        }
        else
        {
            char quote = *s;
            while (*++s && *s != quote)
            {
                if (dos_spec || *s != '\\')
                    /* nothing */;
                else if (s[1] == quote || s[1] == '\\')
                    // From makotom:
                    // This is where the problem happens.
                    // If there are two consecutive `\`s, then the code omits the first one.
                    // In this way `\\` is effectively reduced into `\`.
                    // This is not an expected behaviour, especially when the word is not a file path.
                    s++;
                *p++ = '\\';
                size_t cnt = isascii(*s) ? 1 : mbtowc(NULL, s, MB_CUR_MAX);
                if (cnt <= 1 || cnt == (size_t)-1)
                    *p++ = *s;
                else
                {
                    --s;
                    while (cnt-- > 0)
                        *p++ = *++s;
                }
            }
            if (*s == quote)
                p--;
            if (*s == '\0')
                break;
        }
    }

    *p = '\0';

    printf("intermediate pattern: %s\n", pattern);

    printf("globbed string: %s\n", glob_partial(pattern));
}

int main(void)
{
    // Assume that we are calling `bash -c "echo ' aaa\\\\bbb '"` against Cygwin-based Bash
    // - it is expected to yield a text reading ` aaa\\bbb `, but in fact it does not.
    // We hereby examine argv[2] only to illustrate what is the cause of this unexpected behaviour.
    globify_test("\"echo ' aaa\\\\bbb '\"");
    return 0;
}