"rebase -i": support special-purpose editor to edit insn sheet
[gitweb.git] / convert.c
index 27acce58bc4bec60a394f03db1f6e60e1e4cfc3e..3bb5a4dd57c669bc59be0e2317ef33b64b024992 100644 (file)
--- a/convert.c
+++ b/convert.c
@@ -1,6 +1,7 @@
 #include "cache.h"
 #include "attr.h"
 #include "run-command.h"
+#include "quote.h"
 
 /*
  * convert.c - convert a file when checking it out and checking it in.
@@ -8,13 +9,17 @@
  * This should use the pathname to decide on whether it wants to do some
  * more interesting conversions (automatic gzip/unzip, general format
  * conversions etc etc), but by default it just does automatic CRLF<->LF
- * translation when the "auto_crlf" option is set.
+ * translation when the "text" attribute or "auto_crlf" option is set.
  */
 
-#define CRLF_GUESS     (-1)
-#define CRLF_BINARY    0
-#define CRLF_TEXT      1
-#define CRLF_INPUT     2
+enum crlf_action {
+       CRLF_GUESS = -1,
+       CRLF_BINARY = 0,
+       CRLF_TEXT,
+       CRLF_INPUT,
+       CRLF_CRLF,
+       CRLF_AUTO
+};
 
 struct text_stat {
        /* NUL, CR, LF and CRLF counts */
@@ -89,49 +94,113 @@ static int is_binary(unsigned long size, struct text_stat *stats)
        return 0;
 }
 
-static void check_safe_crlf(const char *path, int action,
+static enum eol output_eol(enum crlf_action crlf_action)
+{
+       switch (crlf_action) {
+       case CRLF_BINARY:
+               return EOL_UNSET;
+       case CRLF_CRLF:
+               return EOL_CRLF;
+       case CRLF_INPUT:
+               return EOL_LF;
+       case CRLF_GUESS:
+               if (!auto_crlf)
+                       return EOL_UNSET;
+               /* fall through */
+       case CRLF_TEXT:
+       case CRLF_AUTO:
+               if (auto_crlf == AUTO_CRLF_TRUE)
+                       return EOL_CRLF;
+               else if (auto_crlf == AUTO_CRLF_INPUT)
+                       return EOL_LF;
+               else if (core_eol == EOL_UNSET)
+                       return EOL_NATIVE;
+       }
+       return core_eol;
+}
+
+static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
                             struct text_stat *stats, enum safe_crlf checksafe)
 {
        if (!checksafe)
                return;
 
-       if (action == CRLF_INPUT || auto_crlf <= 0) {
+       if (output_eol(crlf_action) == EOL_LF) {
                /*
                 * CRLFs would not be restored by checkout:
                 * check if we'd remove CRLFs
                 */
                if (stats->crlf) {
                        if (checksafe == SAFE_CRLF_WARN)
-                               warning("CRLF will be replaced by LF in %s.", path);
+                               warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
                        else /* i.e. SAFE_CRLF_FAIL */
                                die("CRLF would be replaced by LF in %s.", path);
                }
-       } else if (auto_crlf > 0) {
+       } else if (output_eol(crlf_action) == EOL_CRLF) {
                /*
                 * CRLFs would be added by checkout:
                 * check if we have "naked" LFs
                 */
                if (stats->lf != stats->crlf) {
                        if (checksafe == SAFE_CRLF_WARN)
-                               warning("LF will be replaced by CRLF in %s", path);
+                               warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
                        else /* i.e. SAFE_CRLF_FAIL */
                                die("LF would be replaced by CRLF in %s", path);
                }
        }
 }
 
+static int has_cr_in_index(const char *path)
+{
+       int pos, len;
+       unsigned long sz;
+       enum object_type type;
+       void *data;
+       int has_cr;
+       struct index_state *istate = &the_index;
+
+       len = strlen(path);
+       pos = index_name_pos(istate, path, len);
+       if (pos < 0) {
+               /*
+                * We might be in the middle of a merge, in which
+                * case we would read stage #2 (ours).
+                */
+               int i;
+               for (i = -pos - 1;
+                    (pos < 0 && i < istate->cache_nr &&
+                     !strcmp(istate->cache[i]->name, path));
+                    i++)
+                       if (ce_stage(istate->cache[i]) == 2)
+                               pos = i;
+       }
+       if (pos < 0)
+               return 0;
+       data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
+       if (!data || type != OBJ_BLOB) {
+               free(data);
+               return 0;
+       }
+
+       has_cr = memchr(data, '\r', sz) != NULL;
+       free(data);
+       return has_cr;
+}
+
 static int crlf_to_git(const char *path, const char *src, size_t len,
-                       struct strbuf *buf, int action, enum safe_crlf checksafe)
+                      struct strbuf *buf,
+                      enum crlf_action crlf_action, enum safe_crlf checksafe)
 {
        struct text_stat stats;
        char *dst;
 
-       if ((action == CRLF_BINARY) || !auto_crlf || !len)
+       if (crlf_action == CRLF_BINARY ||
+           (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) || !len)
                return 0;
 
        gather_stats(src, len, &stats);
 
-       if (action == CRLF_GUESS) {
+       if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
                /*
                 * We're currently not going to even try to convert stuff
                 * that has bare CR characters. Does anybody do that crazy
@@ -145,9 +214,18 @@ static int crlf_to_git(const char *path, const char *src, size_t len,
                 */
                if (is_binary(len, &stats))
                        return 0;
+
+               if (crlf_action == CRLF_GUESS) {
+                       /*
+                        * If the file in the index has any CR in it, do not convert.
+                        * This is the new safer autocrlf handling.
+                        */
+                       if (has_cr_in_index(path))
+                               return 0;
+               }
        }
 
-       check_safe_crlf(path, action, &stats, checksafe);
+       check_safe_crlf(path, crlf_action, &stats, checksafe);
 
        /* Optimization: No CR? Nothing to convert, regardless. */
        if (!stats.cr)
@@ -157,7 +235,7 @@ static int crlf_to_git(const char *path, const char *src, size_t len,
        if (strbuf_avail(buf) + buf->len < len)
                strbuf_grow(buf, len - buf->len);
        dst = buf->buf;
-       if (action == CRLF_GUESS) {
+       if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
                /*
                 * If we guessed, we already know we rejected a file with
                 * lone CR, and we can strip a CR without looking at what
@@ -180,16 +258,12 @@ static int crlf_to_git(const char *path, const char *src, size_t len,
 }
 
 static int crlf_to_worktree(const char *path, const char *src, size_t len,
-                            struct strbuf *buf, int action)
+                           struct strbuf *buf, enum crlf_action crlf_action)
 {
        char *to_free = NULL;
        struct text_stat stats;
 
-       if ((action == CRLF_BINARY) || (action == CRLF_INPUT) ||
-           auto_crlf <= 0)
-               return 0;
-
-       if (!len)
+       if (!len || output_eol(crlf_action) != EOL_CRLF)
                return 0;
 
        gather_stats(src, len, &stats);
@@ -202,7 +276,14 @@ static int crlf_to_worktree(const char *path, const char *src, size_t len,
        if (stats.lf == stats.crlf)
                return 0;
 
-       if (action == CRLF_GUESS) {
+       if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
+               if (crlf_action == CRLF_GUESS) {
+                       /* If we have any CR or CRLF line endings, we do not touch it */
+                       /* This is the new safer autocrlf-handling */
+                       if (stats.cr > 0 || stats.crlf > 0)
+                               return 0;
+               }
+
                /* If we have any bare CR characters, we're not going to touch it */
                if (stats.cr != stats.crlf)
                        return 0;
@@ -239,9 +320,10 @@ struct filter_params {
        const char *src;
        unsigned long size;
        const char *cmd;
+       const char *path;
 };
 
-static int filter_buffer(int fd, void *data)
+static int filter_buffer(int in, int out, void *data)
 {
        /*
         * Spawn cmd and feed the buffer contents through its stdin.
@@ -249,13 +331,31 @@ static int filter_buffer(int fd, void *data)
        struct child_process child_process;
        struct filter_params *params = (struct filter_params *)data;
        int write_err, status;
-       const char *argv[] = { params->cmd, NULL };
+       const char *argv[] = { NULL, NULL };
+
+       /* apply % substitution to cmd */
+       struct strbuf cmd = STRBUF_INIT;
+       struct strbuf path = STRBUF_INIT;
+       struct strbuf_expand_dict_entry dict[] = {
+               { "f", NULL, },
+               { NULL, NULL, },
+       };
+
+       /* quote the path to preserve spaces, etc. */
+       sq_quote_buf(&path, params->path);
+       dict[0].value = path.buf;
+
+       /* expand all %f with the quoted path */
+       strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
+       strbuf_release(&path);
+
+       argv[0] = cmd.buf;
 
        memset(&child_process, 0, sizeof(child_process));
        child_process.argv = argv;
        child_process.use_shell = 1;
        child_process.in = -1;
-       child_process.out = fd;
+       child_process.out = out;
 
        if (start_command(&child_process))
                return error("cannot fork to run external filter %s", params->cmd);
@@ -269,6 +369,8 @@ static int filter_buffer(int fd, void *data)
        status = finish_command(&child_process);
        if (status)
                error("external filter %s failed %d", params->cmd, status);
+
+       strbuf_release(&cmd);
        return (write_err || status);
 }
 
@@ -292,9 +394,11 @@ static int apply_filter(const char *path, const char *src, size_t len,
        memset(&async, 0, sizeof(async));
        async.proc = filter_buffer;
        async.data = &params;
+       async.out = -1;
        params.src = src;
        params.size = len;
        params.cmd = cmd;
+       params.path = path;
 
        fflush(NULL);
        if (start_async(&async))
@@ -371,24 +475,6 @@ static int read_convert_config(const char *var, const char *value, void *cb)
        return 0;
 }
 
-static void setup_convert_check(struct git_attr_check *check)
-{
-       static struct git_attr *attr_crlf;
-       static struct git_attr *attr_ident;
-       static struct git_attr *attr_filter;
-
-       if (!attr_crlf) {
-               attr_crlf = git_attr("crlf");
-               attr_ident = git_attr("ident");
-               attr_filter = git_attr("filter");
-               user_convert_tail = &user_convert;
-               git_config(read_convert_config, NULL);
-       }
-       check[0].attr = attr_crlf;
-       check[1].attr = attr_ident;
-       check[2].attr = attr_filter;
-}
-
 static int count_ident(const char *cp, unsigned long size)
 {
        /*
@@ -424,6 +510,8 @@ static int count_ident(const char *cp, unsigned long size)
                                cnt++;
                                break;
                        }
+                       if (ch == '\n')
+                               break;
                }
        }
        return cnt;
@@ -445,7 +533,7 @@ static int ident_to_git(const char *path, const char *src, size_t len,
                dollar = memchr(src, '$', len);
                if (!dollar)
                        break;
-               memcpy(dst, src, dollar + 1 - src);
+               memmove(dst, src, dollar + 1 - src);
                dst += dollar + 1 - src;
                len -= dollar + 1 - src;
                src  = dollar + 1;
@@ -454,13 +542,18 @@ static int ident_to_git(const char *path, const char *src, size_t len,
                        dollar = memchr(src + 3, '$', len - 3);
                        if (!dollar)
                                break;
+                       if (memchr(src + 3, '\n', dollar - src - 3)) {
+                               /* Line break before the next dollar. */
+                               continue;
+                       }
+
                        memcpy(dst, "Id$", 3);
                        dst += 3;
                        len -= dollar + 1 - src;
                        src  = dollar + 1;
                }
        }
-       memcpy(dst, src, len);
+       memmove(dst, src, len);
        strbuf_setlen(buf, dst + len - buf->buf);
        return 1;
 }
@@ -469,7 +562,7 @@ static int ident_to_worktree(const char *path, const char *src, size_t len,
                              struct strbuf *buf, int ident)
 {
        unsigned char sha1[20];
-       char *to_free = NULL, *dollar;
+       char *to_free = NULL, *dollar, *spc;
        int cnt;
 
        if (!ident)
@@ -505,7 +598,10 @@ static int ident_to_worktree(const char *path, const char *src, size_t len,
                } else if (src[2] == ':') {
                        /*
                         * It's possible that an expanded Id has crept its way into the
-                        * repository, we cope with that by stripping the expansion out
+                        * repository, we cope with that by stripping the expansion out.
+                        * This is probably not a good idea, since it will cause changes
+                        * on checkout, which won't go away by stash, but let's keep it
+                        * for git-style ids.
                         */
                        dollar = memchr(src + 3, '$', len - 3);
                        if (!dollar) {
@@ -513,6 +609,20 @@ static int ident_to_worktree(const char *path, const char *src, size_t len,
                                break;
                        }
 
+                       if (memchr(src + 3, '\n', dollar - src - 3)) {
+                               /* Line break before the next dollar. */
+                               continue;
+                       }
+
+                       spc = memchr(src + 4, ' ', dollar - src - 4);
+                       if (spc && spc < dollar-1) {
+                               /* There are spaces in unexpected places.
+                                * This is probably an id from some other
+                                * versioning system. Keep it for now.
+                                */
+                               continue;
+                       }
+
                        len -= dollar + 1 - src;
                        src  = dollar + 1;
                } else {
@@ -543,9 +653,24 @@ static int git_path_check_crlf(const char *path, struct git_attr_check *check)
                ;
        else if (!strcmp(value, "input"))
                return CRLF_INPUT;
+       else if (!strcmp(value, "auto"))
+               return CRLF_AUTO;
        return CRLF_GUESS;
 }
 
+static int git_path_check_eol(const char *path, struct git_attr_check *check)
+{
+       const char *value = check->value;
+
+       if (ATTR_UNSET(value))
+               ;
+       else if (!strcmp(value, "lf"))
+               return EOL_LF;
+       else if (!strcmp(value, "crlf"))
+               return EOL_CRLF;
+       return EOL_UNSET;
+}
+
 static struct convert_driver *git_path_check_convert(const char *path,
                                             struct git_attr_check *check)
 {
@@ -567,63 +692,521 @@ static int git_path_check_ident(const char *path, struct git_attr_check *check)
        return !!ATTR_TRUE(value);
 }
 
+static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
+{
+       if (text_attr == CRLF_BINARY)
+               return CRLF_BINARY;
+       if (eol_attr == EOL_LF)
+               return CRLF_INPUT;
+       if (eol_attr == EOL_CRLF)
+               return CRLF_CRLF;
+       return text_attr;
+}
+
+struct conv_attrs {
+       struct convert_driver *drv;
+       enum crlf_action crlf_action;
+       enum eol eol_attr;
+       int ident;
+};
+
+static const char *conv_attr_name[] = {
+       "crlf", "ident", "filter", "eol", "text",
+};
+#define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
+
+static void convert_attrs(struct conv_attrs *ca, const char *path)
+{
+       int i;
+       static struct git_attr_check ccheck[NUM_CONV_ATTRS];
+
+       if (!ccheck[0].attr) {
+               for (i = 0; i < NUM_CONV_ATTRS; i++)
+                       ccheck[i].attr = git_attr(conv_attr_name[i]);
+               user_convert_tail = &user_convert;
+               git_config(read_convert_config, NULL);
+       }
+
+       if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
+               ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
+               if (ca->crlf_action == CRLF_GUESS)
+                       ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
+               ca->ident = git_path_check_ident(path, ccheck + 1);
+               ca->drv = git_path_check_convert(path, ccheck + 2);
+               ca->eol_attr = git_path_check_eol(path, ccheck + 3);
+       } else {
+               ca->drv = NULL;
+               ca->crlf_action = CRLF_GUESS;
+               ca->eol_attr = EOL_UNSET;
+               ca->ident = 0;
+       }
+}
+
 int convert_to_git(const char *path, const char *src, size_t len,
                    struct strbuf *dst, enum safe_crlf checksafe)
 {
-       struct git_attr_check check[3];
-       int crlf = CRLF_GUESS;
-       int ident = 0, ret = 0;
+       int ret = 0;
        const char *filter = NULL;
+       struct conv_attrs ca;
 
-       setup_convert_check(check);
-       if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
-               struct convert_driver *drv;
-               crlf = git_path_check_crlf(path, check + 0);
-               ident = git_path_check_ident(path, check + 1);
-               drv = git_path_check_convert(path, check + 2);
-               if (drv && drv->clean)
-                       filter = drv->clean;
-       }
+       convert_attrs(&ca, path);
+       if (ca.drv)
+               filter = ca.drv->clean;
 
        ret |= apply_filter(path, src, len, dst, filter);
        if (ret) {
                src = dst->buf;
                len = dst->len;
        }
-       ret |= crlf_to_git(path, src, len, dst, crlf, checksafe);
+       ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
+       ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
        if (ret) {
                src = dst->buf;
                len = dst->len;
        }
-       return ret | ident_to_git(path, src, len, dst, ident);
+       return ret | ident_to_git(path, src, len, dst, ca.ident);
 }
 
-int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
+static int convert_to_working_tree_internal(const char *path, const char *src,
+                                           size_t len, struct strbuf *dst,
+                                           int normalizing)
 {
-       struct git_attr_check check[3];
-       int crlf = CRLF_GUESS;
-       int ident = 0, ret = 0;
+       int ret = 0;
        const char *filter = NULL;
+       struct conv_attrs ca;
 
-       setup_convert_check(check);
-       if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
-               struct convert_driver *drv;
-               crlf = git_path_check_crlf(path, check + 0);
-               ident = git_path_check_ident(path, check + 1);
-               drv = git_path_check_convert(path, check + 2);
-               if (drv && drv->smudge)
-                       filter = drv->smudge;
-       }
+       convert_attrs(&ca, path);
+       if (ca.drv)
+               filter = ca.drv->smudge;
 
-       ret |= ident_to_worktree(path, src, len, dst, ident);
+       ret |= ident_to_worktree(path, src, len, dst, ca.ident);
        if (ret) {
                src = dst->buf;
                len = dst->len;
        }
-       ret |= crlf_to_worktree(path, src, len, dst, crlf);
+       /*
+        * CRLF conversion can be skipped if normalizing, unless there
+        * is a smudge filter.  The filter might expect CRLFs.
+        */
+       if (filter || !normalizing) {
+               ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
+               ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
+               if (ret) {
+                       src = dst->buf;
+                       len = dst->len;
+               }
+       }
+       return ret | apply_filter(path, src, len, dst, filter);
+}
+
+int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
+{
+       return convert_to_working_tree_internal(path, src, len, dst, 0);
+}
+
+int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
+{
+       int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
        if (ret) {
                src = dst->buf;
                len = dst->len;
        }
-       return ret | apply_filter(path, src, len, dst, filter);
+       return ret | convert_to_git(path, src, len, dst, 0);
+}
+
+/*****************************************************************
+ *
+ * Streaming converison support
+ *
+ *****************************************************************/
+
+typedef int (*filter_fn)(struct stream_filter *,
+                        const char *input, size_t *isize_p,
+                        char *output, size_t *osize_p);
+typedef void (*free_fn)(struct stream_filter *);
+
+struct stream_filter_vtbl {
+       filter_fn filter;
+       free_fn free;
+};
+
+struct stream_filter {
+       struct stream_filter_vtbl *vtbl;
+};
+
+static int null_filter_fn(struct stream_filter *filter,
+                         const char *input, size_t *isize_p,
+                         char *output, size_t *osize_p)
+{
+       size_t count;
+
+       if (!input)
+               return 0; /* we do not keep any states */
+       count = *isize_p;
+       if (*osize_p < count)
+               count = *osize_p;
+       if (count) {
+               memmove(output, input, count);
+               *isize_p -= count;
+               *osize_p -= count;
+       }
+       return 0;
+}
+
+static void null_free_fn(struct stream_filter *filter)
+{
+       ; /* nothing -- null instances are shared */
+}
+
+static struct stream_filter_vtbl null_vtbl = {
+       null_filter_fn,
+       null_free_fn,
+};
+
+static struct stream_filter null_filter_singleton = {
+       &null_vtbl,
+};
+
+int is_null_stream_filter(struct stream_filter *filter)
+{
+       return filter == &null_filter_singleton;
+}
+
+
+/*
+ * LF-to-CRLF filter
+ */
+static int lf_to_crlf_filter_fn(struct stream_filter *filter,
+                               const char *input, size_t *isize_p,
+                               char *output, size_t *osize_p)
+{
+       size_t count;
+
+       if (!input)
+               return 0; /* we do not keep any states */
+       count = *isize_p;
+       if (count) {
+               size_t i, o;
+               for (i = o = 0; o < *osize_p && i < count; i++) {
+                       char ch = input[i];
+                       if (ch == '\n') {
+                               if (o + 1 < *osize_p)
+                                       output[o++] = '\r';
+                               else
+                                       break;
+                       }
+                       output[o++] = ch;
+               }
+
+               *osize_p -= o;
+               *isize_p -= i;
+       }
+       return 0;
+}
+
+static struct stream_filter_vtbl lf_to_crlf_vtbl = {
+       lf_to_crlf_filter_fn,
+       null_free_fn,
+};
+
+static struct stream_filter lf_to_crlf_filter_singleton = {
+       &lf_to_crlf_vtbl,
+};
+
+
+/*
+ * Cascade filter
+ */
+#define FILTER_BUFFER 1024
+struct cascade_filter {
+       struct stream_filter filter;
+       struct stream_filter *one;
+       struct stream_filter *two;
+       char buf[FILTER_BUFFER];
+       int end, ptr;
+};
+
+static int cascade_filter_fn(struct stream_filter *filter,
+                            const char *input, size_t *isize_p,
+                            char *output, size_t *osize_p)
+{
+       struct cascade_filter *cas = (struct cascade_filter *) filter;
+       size_t filled = 0;
+       size_t sz = *osize_p;
+       size_t to_feed, remaining;
+
+       /*
+        * input -- (one) --> buf -- (two) --> output
+        */
+       while (filled < sz) {
+               remaining = sz - filled;
+
+               /* do we already have something to feed two with? */
+               if (cas->ptr < cas->end) {
+                       to_feed = cas->end - cas->ptr;
+                       if (stream_filter(cas->two,
+                                         cas->buf + cas->ptr, &to_feed,
+                                         output + filled, &remaining))
+                               return -1;
+                       cas->ptr += (cas->end - cas->ptr) - to_feed;
+                       filled = sz - remaining;
+                       continue;
+               }
+
+               /* feed one from upstream and have it emit into our buffer */
+               to_feed = input ? *isize_p : 0;
+               if (input && !to_feed)
+                       break;
+               remaining = sizeof(cas->buf);
+               if (stream_filter(cas->one,
+                                 input, &to_feed,
+                                 cas->buf, &remaining))
+                       return -1;
+               cas->end = sizeof(cas->buf) - remaining;
+               cas->ptr = 0;
+               if (input) {
+                       size_t fed = *isize_p - to_feed;
+                       *isize_p -= fed;
+                       input += fed;
+               }
+
+               /* do we know that we drained one completely? */
+               if (input || cas->end)
+                       continue;
+
+               /* tell two to drain; we have nothing more to give it */
+               to_feed = 0;
+               remaining = sz - filled;
+               if (stream_filter(cas->two,
+                                 NULL, &to_feed,
+                                 output + filled, &remaining))
+                       return -1;
+               if (remaining == (sz - filled))
+                       break; /* completely drained two */
+               filled = sz - remaining;
+       }
+       *osize_p -= filled;
+       return 0;
+}
+
+static void cascade_free_fn(struct stream_filter *filter)
+{
+       struct cascade_filter *cas = (struct cascade_filter *)filter;
+       free_stream_filter(cas->one);
+       free_stream_filter(cas->two);
+       free(filter);
+}
+
+static struct stream_filter_vtbl cascade_vtbl = {
+       cascade_filter_fn,
+       cascade_free_fn,
+};
+
+static struct stream_filter *cascade_filter(struct stream_filter *one,
+                                           struct stream_filter *two)
+{
+       struct cascade_filter *cascade;
+
+       if (!one || is_null_stream_filter(one))
+               return two;
+       if (!two || is_null_stream_filter(two))
+               return one;
+
+       cascade = xmalloc(sizeof(*cascade));
+       cascade->one = one;
+       cascade->two = two;
+       cascade->end = cascade->ptr = 0;
+       cascade->filter.vtbl = &cascade_vtbl;
+       return (struct stream_filter *)cascade;
+}
+
+/*
+ * ident filter
+ */
+#define IDENT_DRAINING (-1)
+#define IDENT_SKIPPING (-2)
+struct ident_filter {
+       struct stream_filter filter;
+       struct strbuf left;
+       int state;
+       char ident[45]; /* ": x40 $" */
+};
+
+static int is_foreign_ident(const char *str)
+{
+       int i;
+
+       if (prefixcmp(str, "$Id: "))
+               return 0;
+       for (i = 5; str[i]; i++) {
+               if (isspace(str[i]) && str[i+1] != '$')
+                       return 1;
+       }
+       return 0;
+}
+
+static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
+{
+       size_t to_drain = ident->left.len;
+
+       if (*osize_p < to_drain)
+               to_drain = *osize_p;
+       if (to_drain) {
+               memcpy(*output_p, ident->left.buf, to_drain);
+               strbuf_remove(&ident->left, 0, to_drain);
+               *output_p += to_drain;
+               *osize_p -= to_drain;
+       }
+       if (!ident->left.len)
+               ident->state = 0;
+}
+
+static int ident_filter_fn(struct stream_filter *filter,
+                          const char *input, size_t *isize_p,
+                          char *output, size_t *osize_p)
+{
+       struct ident_filter *ident = (struct ident_filter *)filter;
+       static const char head[] = "$Id";
+
+       if (!input) {
+               /* drain upon eof */
+               switch (ident->state) {
+               default:
+                       strbuf_add(&ident->left, head, ident->state);
+               case IDENT_SKIPPING:
+                       /* fallthru */
+               case IDENT_DRAINING:
+                       ident_drain(ident, &output, osize_p);
+               }
+               return 0;
+       }
+
+       while (*isize_p || (ident->state == IDENT_DRAINING)) {
+               int ch;
+
+               if (ident->state == IDENT_DRAINING) {
+                       ident_drain(ident, &output, osize_p);
+                       if (!*osize_p)
+                               break;
+                       continue;
+               }
+
+               ch = *(input++);
+               (*isize_p)--;
+
+               if (ident->state == IDENT_SKIPPING) {
+                       /*
+                        * Skipping until '$' or LF, but keeping them
+                        * in case it is a foreign ident.
+                        */
+                       strbuf_addch(&ident->left, ch);
+                       if (ch != '\n' && ch != '$')
+                               continue;
+                       if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
+                               strbuf_setlen(&ident->left, sizeof(head) - 1);
+                               strbuf_addstr(&ident->left, ident->ident);
+                       }
+                       ident->state = IDENT_DRAINING;
+                       continue;
+               }
+
+               if (ident->state < sizeof(head) &&
+                   head[ident->state] == ch) {
+                       ident->state++;
+                       continue;
+               }
+
+               if (ident->state)
+                       strbuf_add(&ident->left, head, ident->state);
+               if (ident->state == sizeof(head) - 1) {
+                       if (ch != ':' && ch != '$') {
+                               strbuf_addch(&ident->left, ch);
+                               ident->state = 0;
+                               continue;
+                       }
+
+                       if (ch == ':') {
+                               strbuf_addch(&ident->left, ch);
+                               ident->state = IDENT_SKIPPING;
+                       } else {
+                               strbuf_addstr(&ident->left, ident->ident);
+                               ident->state = IDENT_DRAINING;
+                       }
+                       continue;
+               }
+
+               strbuf_addch(&ident->left, ch);
+               ident->state = IDENT_DRAINING;
+       }
+       return 0;
+}
+
+static void ident_free_fn(struct stream_filter *filter)
+{
+       struct ident_filter *ident = (struct ident_filter *)filter;
+       strbuf_release(&ident->left);
+       free(filter);
+}
+
+static struct stream_filter_vtbl ident_vtbl = {
+       ident_filter_fn,
+       ident_free_fn,
+};
+
+static struct stream_filter *ident_filter(const unsigned char *sha1)
+{
+       struct ident_filter *ident = xmalloc(sizeof(*ident));
+
+       sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
+       strbuf_init(&ident->left, 0);
+       ident->filter.vtbl = &ident_vtbl;
+       ident->state = 0;
+       return (struct stream_filter *)ident;
+}
+
+/*
+ * Return an appropriately constructed filter for the path, or NULL if
+ * the contents cannot be filtered without reading the whole thing
+ * in-core.
+ *
+ * Note that you would be crazy to set CRLF, smuge/clean or ident to a
+ * large binary blob you would want us not to slurp into the memory!
+ */
+struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
+{
+       struct conv_attrs ca;
+       enum crlf_action crlf_action;
+       struct stream_filter *filter = NULL;
+
+       convert_attrs(&ca, path);
+
+       if (ca.drv && (ca.drv->smudge || ca.drv->clean))
+               return filter;
+
+       if (ca.ident)
+               filter = ident_filter(sha1);
+
+       crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
+
+       if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
+           (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
+               filter = cascade_filter(filter, &null_filter_singleton);
+
+       else if (output_eol(crlf_action) == EOL_CRLF &&
+                !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
+               filter = cascade_filter(filter, &lf_to_crlf_filter_singleton);
+
+       return filter;
+}
+
+void free_stream_filter(struct stream_filter *filter)
+{
+       filter->vtbl->free(filter);
+}
+
+int stream_filter(struct stream_filter *filter,
+                 const char *input, size_t *isize_p,
+                 char *output, size_t *osize_p)
+{
+       return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
 }